From a5251ca11d2317d93a7b6da4217483f4e83beb3d Mon Sep 17 00:00:00 2001
From: "Piotr Wilkin (ilintar)" <piotr.wilkin@syndatis.com>
Date: Tue, 16 Dec 2025 11:59:53 +0100
Subject: [PATCH 1/6] Optimization: Qwen3 next autoregressive pass (#17996)

* It's Qwen3 Next, the lean mean token generation machine!

* Apply patches from thread

* Remove recurrent version, only keep chunked and autoregressive

* Remove unnecessary conts and asserts

* Remove more extra conts and asserts

* Cleanup masking
---
 src/models/models.h      |  22 +--
 src/models/qwen3next.cpp | 333 +++++++++------------------------------
 2 files changed, 85 insertions(+), 270 deletions(-)

diff --git a/src/models/models.h b/src/models/models.h
index 6494f54501..ffb36acc61 100644
--- a/src/models/models.h
+++ b/src/models/models.h
@@ -441,23 +441,13 @@ private:
                 ggml_tensor * cur,
                 ggml_tensor * causal_mask,
                 ggml_tensor * identity,
+                ggml_tensor * diag_mask,
                         int   il);
 
     ggml_tensor * build_layer_ffn(
                 ggml_tensor * cur,
                         int   il);
 
-    ggml_tensor * build_delta_net_recurrent(
-                ggml_tensor * q,
-                ggml_tensor * k,
-                ggml_tensor * v,
-                ggml_tensor * g,
-                ggml_tensor * beta,
-                ggml_tensor * state,
-                ggml_tensor * causal_mask,
-                ggml_tensor * identity,
-                        int   il);
-
     ggml_tensor * build_delta_net_chunking(
                 ggml_tensor * q,
                 ggml_tensor * k,
@@ -467,8 +457,18 @@ private:
                 ggml_tensor * state,
                 ggml_tensor * causal_mask,
                 ggml_tensor * identity,
+                ggml_tensor * diag_mask,
                         int   il);
 
+    ggml_tensor * build_delta_net_autoregressive(
+                ggml_tensor * q,
+                ggml_tensor * k,
+                ggml_tensor * v,
+                ggml_tensor * g,
+                ggml_tensor * beta,
+                ggml_tensor * state,
+                int           il);
+
     ggml_tensor * build_norm_gated(
                 ggml_tensor * input,
                 ggml_tensor * weights,
diff --git a/src/models/qwen3next.cpp b/src/models/qwen3next.cpp
index c8f1b5ec90..775b3135d3 100644
--- a/src/models/qwen3next.cpp
+++ b/src/models/qwen3next.cpp
@@ -17,13 +17,15 @@ llm_build_qwen3next::llm_build_qwen3next(const llama_model & model, const llm_gr
     ggml_tensor * inp_out_ids = build_inp_out_ids();
 
     ggml_tensor * causal_mask =
-        ggml_tri(ctx0, ggml_fill_inplace(ctx0, ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, ubatch.n_seq_tokens, ubatch.n_seq_tokens), 1.0f),
+        ggml_tri(ctx0, ggml_fill_inplace(ctx0, ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, CHUNK_SIZE, CHUNK_SIZE), 1.0f),
                     GGML_TRI_TYPE_LOWER);
 
-    ggml_tensor * identity = ggml_diag(ctx0, ggml_fill_inplace(ctx0, ggml_new_tensor_1d(ctx0, GGML_TYPE_F32, ubatch.n_seq_tokens), 1.0f));
+    ggml_tensor * identity = ggml_diag(ctx0, ggml_fill_inplace(ctx0, ggml_new_tensor_1d(ctx0, GGML_TYPE_F32, CHUNK_SIZE), 1.0f));
+    ggml_tensor * diag_mask = ggml_add(ctx0, causal_mask, identity);
 
     ggml_build_forward_expand(gf, causal_mask);
     ggml_build_forward_expand(gf, identity);
+    ggml_build_forward_expand(gf, diag_mask);
 
     for (int il = 0; il < n_layer; ++il) {
         ggml_tensor * inpSA = inpL;
@@ -34,7 +36,7 @@ llm_build_qwen3next::llm_build_qwen3next(const llama_model & model, const llm_gr
         // Determine layer type and build appropriate attention mechanism
         if (hparams.is_recurrent(il)) {
             // Linear attention layer (gated delta net)
-            cur = build_layer_attn_linear(inp->get_recr(), cur, causal_mask, identity, il);
+            cur = build_layer_attn_linear(inp->get_recr(), cur, causal_mask, identity, diag_mask, il);
         } else {
             // Full attention layer
             cur = build_layer_attn(inp->get_attn(), cur, inp_pos, il);
@@ -93,14 +95,8 @@ ggml_tensor * llm_build_qwen3next::build_delta_net_chunking(
         ggml_tensor * state,
         ggml_tensor * causal_mask,
         ggml_tensor * identity,
+        ggml_tensor * diag_mask,
         int           il) {
-    GGML_ASSERT(ggml_is_contiguous(q));
-    GGML_ASSERT(ggml_is_contiguous(k));
-    GGML_ASSERT(ggml_is_contiguous(v));
-    GGML_ASSERT(ggml_is_contiguous(g));
-    GGML_ASSERT(ggml_is_contiguous(beta));
-    GGML_ASSERT(ggml_is_contiguous(state));
-
     const int64_t S_k      = q->ne[0];
     const int64_t H_k      = q->ne[1];
     const int64_t n_tokens = q->ne[2];
@@ -120,15 +116,10 @@ ggml_tensor * llm_build_qwen3next::build_delta_net_chunking(
 
     GGML_ASSERT(H_k == H_v);  // we did a repeat to make sure this is the case
 
-    // TODO: can this ever be false?
-    const bool use_qk_l2norm = true;
+    const float eps_norm = hparams.f_norm_rms_eps;
 
-    if (use_qk_l2norm) {
-        const float eps_norm = hparams.f_norm_rms_eps;
-
-        q = ggml_l2_norm(ctx0, q, eps_norm);
-        k = ggml_l2_norm(ctx0, k, eps_norm);
-    }
+    q = ggml_l2_norm(ctx0, q, eps_norm);
+    k = ggml_l2_norm(ctx0, k, eps_norm);
 
     const float scale = 1.0f / sqrtf(S_v);
 
@@ -136,8 +127,6 @@ ggml_tensor * llm_build_qwen3next::build_delta_net_chunking(
 
     beta = ggml_sigmoid(ctx0, beta);
 
-    ggml_tensor * causal_diag_mask = ggml_add(ctx0, causal_mask, identity);
-
     cb(q, "q_in", il);
     cb(k, "k_in", il);
     cb(v, "v_in", il);
@@ -188,36 +177,21 @@ ggml_tensor * llm_build_qwen3next::build_delta_net_chunking(
     cb(v_beta, "v_beta", il);
     cb(k_beta, "k_beta", il);
 
-    ggml_tensor * chunked_mask =
-        ggml_view_4d(ctx0, causal_mask, chunk_size,
-                chunk_size,         causal_mask->ne[2], causal_mask->ne[3],
-                causal_mask->nb[1], causal_mask->nb[2], causal_mask->nb[3], 0);
+    q      = ggml_reshape_4d(ctx0, q,      S_k, chunk_size, n_chunks, H_k * n_seqs);
+    k      = ggml_reshape_4d(ctx0, k,      S_k, chunk_size, n_chunks, H_k * n_seqs);
+    k_beta = ggml_reshape_4d(ctx0, k_beta, S_k, chunk_size, n_chunks, H_k * n_seqs);
+    v      = ggml_reshape_4d(ctx0, v,      S_v, chunk_size, n_chunks, H_v * n_seqs);
+    v_beta = ggml_reshape_4d(ctx0, v_beta, S_v, chunk_size, n_chunks, H_v * n_seqs);
 
-    ggml_tensor * chunked_diag_mask =
-        ggml_view_4d(ctx0, causal_diag_mask, chunk_size,
-                chunk_size,              causal_diag_mask->ne[2], causal_diag_mask->ne[3],
-                causal_diag_mask->nb[1], causal_diag_mask->nb[2], causal_diag_mask->nb[3], 0);
-
-    ggml_tensor * chunked_identity =
-        ggml_view_4d(ctx0, identity, chunk_size,
-            chunk_size,      identity->ne[2], identity->ne[3],
-            identity->nb[1], identity->nb[2], identity->nb[3], 0);
-
-    q      = ggml_cont_4d(ctx0, q,      S_k, chunk_size, n_chunks, H_k * n_seqs);
-    k      = ggml_cont_4d(ctx0, k,      S_k, chunk_size, n_chunks, H_k * n_seqs);
-    k_beta = ggml_cont_4d(ctx0, k_beta, S_k, chunk_size, n_chunks, H_k * n_seqs);
-    v      = ggml_cont_4d(ctx0, v,      S_v, chunk_size, n_chunks, H_v * n_seqs);
-    v_beta = ggml_cont_4d(ctx0, v_beta, S_v, chunk_size, n_chunks, H_v * n_seqs);
-
-    g    = ggml_cont_4d(ctx0, g, chunk_size, 1, n_chunks, H_k * n_seqs);
-    beta = ggml_cont_4d(ctx0, beta, 1, chunk_size, n_chunks, H_k * n_seqs);
+    g    = ggml_reshape_4d(ctx0, g, chunk_size, 1, n_chunks, H_k * n_seqs);
+    beta = ggml_reshape_4d(ctx0, beta, 1, chunk_size, n_chunks, H_k * n_seqs);
 
     ggml_tensor * g_cumsum = ggml_cumsum(ctx0, g);
 
     cb(g_cumsum, "g_cumsum", il);
 
-    ggml_tensor * gcs_i = ggml_cont_4d(ctx0, g_cumsum, chunk_size, 1, n_chunks, H_v * n_seqs);
-    ggml_tensor * gcs_j = ggml_cont_4d(ctx0, g_cumsum, 1, chunk_size, n_chunks, H_v * n_seqs);
+    ggml_tensor * gcs_i = ggml_reshape_4d(ctx0, g_cumsum, chunk_size, 1, n_chunks, H_v * n_seqs);
+    ggml_tensor * gcs_j = ggml_reshape_4d(ctx0, g_cumsum, 1, chunk_size, n_chunks, H_v * n_seqs);
 
     ggml_tensor * gcs_j_broadcast =
         ggml_repeat_4d(ctx0, gcs_j, chunk_size, chunk_size, n_chunks, H_v * n_seqs);
@@ -226,23 +200,23 @@ ggml_tensor * llm_build_qwen3next::build_delta_net_chunking(
 
     cb(decay_mask, "decay_mask", il);
 
-    decay_mask = ggml_mul(ctx0, decay_mask, chunked_diag_mask);
+    decay_mask = ggml_mul(ctx0, decay_mask, diag_mask);
     decay_mask = ggml_exp(ctx0, decay_mask);
-    decay_mask = ggml_mul(ctx0, decay_mask, chunked_diag_mask);
+    decay_mask = ggml_mul(ctx0, decay_mask, diag_mask);
 
     ggml_tensor * kmulkbeta = ggml_mul_mat(ctx0, k, k_beta);
 
     ggml_tensor * k_decay = ggml_mul(ctx0, kmulkbeta, decay_mask);
-    ggml_tensor * attn    = ggml_neg(ctx0, ggml_mul(ctx0, k_decay, chunked_mask));
+    ggml_tensor * attn    = ggml_neg(ctx0, ggml_mul(ctx0, k_decay, causal_mask));
 
     cb(attn, "attn_pre_solve", il);
 
-    ggml_tensor * attn_lower = ggml_mul(ctx0, attn, chunked_mask);
-    ggml_tensor * lhs        = ggml_sub(ctx0, ggml_repeat(ctx0, chunked_identity, attn_lower), attn_lower);
+    ggml_tensor * attn_lower = ggml_mul(ctx0, attn, causal_mask);
+    ggml_tensor * lhs        = ggml_sub(ctx0, ggml_repeat(ctx0, identity, attn_lower), attn_lower);
 
     ggml_tensor * lin_solve  = ggml_solve_tri(ctx0, lhs, attn, true, true, false);
-    attn                     = ggml_mul(ctx0, lin_solve, chunked_mask);
-    attn                     = ggml_add(ctx0, attn, chunked_identity);
+    attn                     = ggml_mul(ctx0, lin_solve, causal_mask);
+    attn                     = ggml_add(ctx0, attn, identity);
 
     cb(attn, "attn_solved", il);
 
@@ -291,7 +265,7 @@ ggml_tensor * llm_build_qwen3next::build_delta_net_chunking(
         // attn = (q_i @ k_i.transpose(-1, -2) * decay_mask[:, :, i]).masked_fill_(mask, 0)
         attn = ggml_mul_mat(ctx0, k_chunk, q_chunk);
         attn = ggml_mul(ctx0, attn, decay_mask_chunk);
-        attn = ggml_mul(ctx0, attn, ggml_add(ctx0, chunked_identity, chunked_mask));
+        attn = ggml_mul(ctx0, attn, diag_mask);
 
         ggml_tensor * state_t = ggml_cont_4d(ctx0, ggml_permute(ctx0, new_state, 1, 0, 2, 3), S_v, S_v, 1, H_v * n_seqs);
 
@@ -361,23 +335,14 @@ ggml_tensor * llm_build_qwen3next::build_delta_net_chunking(
     return ggml_concat(ctx0, flat_output, flat_state, 0);
 }
 
-ggml_tensor * llm_build_qwen3next::build_delta_net_recurrent(
+ggml_tensor * llm_build_qwen3next::build_delta_net_autoregressive(
         ggml_tensor * q,
         ggml_tensor * k,
         ggml_tensor * v,
         ggml_tensor * g,
         ggml_tensor * beta,
         ggml_tensor * state,
-        ggml_tensor * causal_mask,
-        ggml_tensor * identity,
         int           il) {
-    GGML_ASSERT(ggml_is_contiguous(q));
-    GGML_ASSERT(ggml_is_contiguous(k));
-    GGML_ASSERT(ggml_is_contiguous(v));
-    GGML_ASSERT(ggml_is_contiguous(g));
-    GGML_ASSERT(ggml_is_contiguous(beta));
-    GGML_ASSERT(ggml_is_contiguous(state));
-
     const int64_t S_k      = q->ne[0];
     const int64_t H_k      = q->ne[1];
     const int64_t n_tokens = q->ne[2];
@@ -386,6 +351,7 @@ ggml_tensor * llm_build_qwen3next::build_delta_net_recurrent(
     const int64_t S_v = v->ne[0];
     const int64_t H_v = v->ne[1];
 
+    GGML_ASSERT(n_tokens == 1);  // This function is optimized for single token processing
     GGML_ASSERT(v->ne[2] == n_tokens);
     GGML_ASSERT(k->ne[2] == n_tokens);
     GGML_ASSERT(g->ne[0] == H_v && g->ne[1] == n_tokens && g->ne[2] == n_seqs);
@@ -397,215 +363,65 @@ ggml_tensor * llm_build_qwen3next::build_delta_net_recurrent(
 
     GGML_ASSERT(H_k == H_v);  // we did a repeat to make sure this is the case
 
-    // TODO: can this ever be false?
-    const bool use_qk_l2norm = true;
+    const float eps_norm = hparams.f_norm_rms_eps;
 
-    if (use_qk_l2norm) {
-        const float eps_norm = hparams.f_norm_rms_eps;
-
-        q = ggml_l2_norm(ctx0, q, eps_norm);
-        k = ggml_l2_norm(ctx0, k, eps_norm);
-    }
+    q = ggml_l2_norm(ctx0, q, eps_norm);
+    k = ggml_l2_norm(ctx0, k, eps_norm);
 
     const float scale = 1.0f / sqrtf(S_v);
 
-    q = ggml_scale(ctx0, q, scale);
-
+    q    = ggml_scale(ctx0, q, scale);
     beta = ggml_sigmoid(ctx0, beta);
 
-    ggml_tensor * causal_diag_mask = ggml_add(ctx0, causal_mask, identity);
-
     cb(q, "q_in", il);
     cb(k, "k_in", il);
     cb(v, "v_in", il);
     cb(beta, "beta_in", il);
     cb(g, "g_in", il);
 
-    q = ggml_cont_4d(ctx0, ggml_permute(ctx0, q, 0, 2, 1, 3), S_v, n_tokens, H_v, n_seqs);
-    k = ggml_cont_4d(ctx0, ggml_permute(ctx0, k, 0, 2, 1, 3), S_v, n_tokens, H_v, n_seqs);
-    v = ggml_cont_4d(ctx0, ggml_permute(ctx0, v, 0, 2, 1, 3), S_v, n_tokens, H_v, n_seqs);
-    g = ggml_cont_4d(ctx0, ggml_permute(ctx0, g, 2, 0, 3, 1), n_tokens, 1, H_k, n_seqs);
-
-    beta  = ggml_cont(ctx0, ggml_permute(ctx0, beta, 2, 0, 1, 3));
     state = ggml_reshape_4d(ctx0, state, S_v, S_v, H_v, n_seqs);
 
-    cb(q, "q_perm", il);
-    cb(k, "k_perm", il);
-    cb(v, "v_perm", il);
-    cb(beta, "beta_perm", il);
-    cb(g, "g_perm", il);
-    cb(state, "state_in", il);
+    ggml_tensor * g_t    = ggml_reshape_4d(ctx0, ggml_transpose(ctx0, g), 1, 1, H_k, n_seqs);
+    ggml_tensor * beta_t = ggml_reshape_4d(ctx0, ggml_transpose(ctx0, beta), 1, 1, H_k, n_seqs);
 
-    GGML_ASSERT(q->ne[1] == n_tokens && q->ne[0] == S_k && q->ne[2] == H_k && q->ne[3] == n_seqs);
-    GGML_ASSERT(k->ne[1] == n_tokens && k->ne[0] == S_k && k->ne[2] == H_k && k->ne[3] == n_seqs);
-    GGML_ASSERT(v->ne[1] == n_tokens && v->ne[0] == S_v && v->ne[2] == H_k && v->ne[3] == n_seqs);
-    GGML_ASSERT(beta->ne[1] == n_tokens && beta->ne[2] == H_k && beta->ne[0] == 1 && beta->ne[3] == n_seqs);
+    // Apply exponential to g_t
+    g_t = ggml_exp(ctx0, g_t);
 
-    ggml_tensor * v_beta = ggml_mul(ctx0, v, beta);
-    ggml_tensor * k_beta = ggml_mul(ctx0, k, beta);
+    // Apply the gated delta rule for the single timestep
+    // last_recurrent_state = last_recurrent_state * g_t
+    state = ggml_mul(ctx0, state, g_t);
 
-    ggml_tensor * g_cumsum = ggml_cumsum(ctx0, g);
+    // kv_mem = (last_recurrent_state * k_t.unsqueeze(-1)).sum(dim=-2)
+    ggml_tensor * k_t_unsqueezed = ggml_reshape_4d(ctx0, k, 1, S_v, H_v, n_seqs);
+    ggml_tensor * kv_mem         = ggml_mul(ctx0, state, k_t_unsqueezed);
+    // we need to sum over dim=-2, so we transpose, sum, then transpose again
+    kv_mem = ggml_transpose(ctx0, ggml_sum_rows(ctx0, ggml_cont(ctx0, ggml_transpose(ctx0, kv_mem))));
 
-    cb(k_beta, "k_beta", il);
-    cb(v_beta, "v_beta", il);
-    cb(g_cumsum, "g_cumsum", il);
+    // v_t = v.unsqueeze(2) (we insert the singleton dimension after n_seqs and H_v)
+    ggml_tensor * v_t    = ggml_reshape_4d(ctx0, v, S_v, 1, H_v, n_seqs);
+    // delta = (v_t - kv_mem) * beta_t
+    ggml_tensor * v_diff = ggml_sub(ctx0, v_t, kv_mem);  // both should be [S_v, 1, H_v, n_seqs]
+    ggml_tensor * delta  = ggml_mul(ctx0, v_diff, beta_t);
 
-    ggml_tensor * gcs_i = ggml_cont_4d(ctx0, g_cumsum, n_tokens, 1, H_v, n_seqs);  // [chunk_size, 1, n_tokens, n_seqs]
-    ggml_tensor * gcs_j = ggml_cont_4d(ctx0, g_cumsum, 1, n_tokens, H_v, n_seqs);  // [1, chunk_size, n_tokens, n_seqs]
+    // last_recurrent_state = last_recurrent_state + k_t.unsqueeze(-1) * delta
+    ggml_tensor * k_t_delta = ggml_mul(ctx0, ggml_repeat_4d(ctx0, k_t_unsqueezed, S_v, S_v, H_v, n_seqs), delta);
+    state                   = ggml_add(ctx0, state, k_t_delta);
 
-    // Broadcast both tensors to [chunk_size, chunk_size, H_v, n_seqs]
-    // ggml_tensor * gcs_i_broadcast =
-    //     ggml_repeat_4d(ctx0, gcs_i, GGML_DELTA_NET_CHUNK, GGML_DELTA_NET_CHUNK, num_chunks * H_v,
-    //                     n_seqs);  // [chunk_size, 1, H_v, n_seqs] -> [chunk_size, chunk_size, H_v, n_seqs]
-    // Don't need this, this one will get auto-broadcast
-    ggml_tensor * gcs_j_broadcast =
-        ggml_repeat_4d(ctx0, gcs_j, n_tokens, n_tokens, H_v, n_seqs);  // [1, chunk_size, H_v, n_seqs] -> [chunk_size, chunk_size, H_v, n_seqs]
-
-    ggml_tensor * decay_mask = ggml_sub(ctx0, gcs_j_broadcast, gcs_i);
-
-    // Apply lower triangular mask to ensure attention is causal (only past tokens influence current)
-    decay_mask = ggml_mul(ctx0, decay_mask, causal_diag_mask);
-    // Apply exponential to get the decay mask values
-    decay_mask = ggml_exp(ctx0, decay_mask);
-    // Apply lower triangular mask again to ensure only lower triangular values remain
-    decay_mask = ggml_mul(ctx0, decay_mask, causal_diag_mask);
-
-    cb(decay_mask, "decay_mask", il);
-
-    // attn = -((k_beta @ key.transpose(-1, -2)) * decay_mask).masked_fill(mask, 0)
-    ggml_tensor * kmulkbeta = ggml_mul_mat(ctx0, k, k_beta);
-
-    cb(kmulkbeta, "kmulkbeta", il);
-
-    ggml_tensor * k_decay = ggml_mul(ctx0, kmulkbeta, decay_mask);
-    ggml_tensor * attn    = ggml_neg(ctx0, ggml_mul(ctx0, k_decay, causal_mask));
-
-    cb(attn, "attn_pre_rec", il);
-
-    // for i in range(1, chunk_size):
-    //          row = attn[..., i, :i].clone()
-    //          sub = attn[..., :i, :i].clone()
-    //          attn[..., i, :i] = row + (row.unsqueeze(-1) * sub).sum(-2)
-    // attn = attn + torch.eye(chunk_size, dtype=attn.dtype, device=attn.device)
-    //
-    // We reduce this to a linear triangular solve: AX = B, where B = attn, A = I - tril(A)
-    ggml_tensor * attn_lower = ggml_mul(ctx0, attn, causal_mask);
-    ggml_tensor * lhs        = ggml_sub(ctx0, ggml_repeat(ctx0, identity, attn_lower), attn_lower);
-
-    ggml_tensor * lin_solve  = ggml_solve_tri(ctx0, lhs, attn, true, true, false);
-    attn                     = ggml_mul(ctx0, lin_solve, causal_mask);
-    attn                     = ggml_add(ctx0, attn, identity);
-
-    // value = attn @ v_beta
-    v = ggml_mul_mat(ctx0, ggml_cont(ctx0, ggml_transpose(ctx0, v_beta)), attn);
-
-    cb(v, "value_beta", il);
-
-    // k_cumdecay = attn @ (k_beta * g.exp().unsqueeze(-1))
-    ggml_tensor * g_cumsum_t = ggml_cont(ctx0, ggml_transpose(ctx0, g_cumsum));
-    ggml_tensor * gexp       = ggml_exp(ctx0, g_cumsum_t);
-
-    cb(gexp, "g_cum_exp", il);
-
-    ggml_tensor * kbeta_gexp = ggml_mul(ctx0, k_beta, gexp);
-
-    cb(kbeta_gexp, "kbeta_gexp", il);
-
-    ggml_tensor * k_cumdecay =
-        ggml_cont(ctx0, ggml_transpose(ctx0, ggml_mul_mat(ctx0, attn, ggml_cont(ctx0, ggml_transpose(ctx0, kbeta_gexp)))));
-
-    cb(k_cumdecay, "k_cumdecay", il);
-
-    // attn = (q_i @ k_i.transpose(-1, -2) * decay_mask[:, :, i]).masked_fill_(mask, 0)
-    attn = ggml_mul_mat(ctx0, k, q);
-    attn = ggml_mul(ctx0, attn, decay_mask);
-    attn = ggml_mul(ctx0, attn, ggml_add(ctx0, identity, causal_mask));
-
-    cb(attn, "attn_decay_key", il);
-
-    ggml_tensor * state_t = ggml_cont(ctx0, ggml_transpose(ctx0, state));
-
-    // v_prime = (k_cumdecay[:, :, i]) @ last_recurrent_state
-    ggml_tensor * v_prime = ggml_mul_mat(ctx0, state_t, k_cumdecay);
-
-    cb(v_prime, "v_prime", il);
-
-    // v_new = v_i - v_prime
-    ggml_tensor * v_new = ggml_sub(ctx0, ggml_repeat(ctx0, v, v_prime), v_prime);
-
-    ggml_tensor * v_new_t = ggml_cont(ctx0, ggml_transpose(ctx0, v_new));
-
-    cb(v_new, "v_new", il);
-
-    // attn_inter = (q_i * g[:, :, i, :, None].exp()) @ last_recurrent_state
-    ggml_tensor * q_g_exp    = ggml_mul(ctx0, q, gexp);
-    ggml_tensor * attn_inter = ggml_mul_mat(ctx0, state_t, q_g_exp);
-
-    cb(attn_inter, "attn_inter", il);
-
-    // core_attn_out[:, :, i] = attn_inter + attn @ v_new
-    ggml_tensor * v_attn = ggml_mul_mat(ctx0, v_new_t, attn);
-
-    cb(v_attn, "v_attn", il);
-
-    ggml_tensor * core_attn_out = ggml_add(ctx0, attn_inter, v_attn);
-
-    cb(core_attn_out, "core_attn_out", il);
-
-    // g_last = torch.clamp(g_cum[:, :, -1], max=50.0).exp().unsqueeze(-1).unsqueeze(-1)
-    // g_diff = torch.clamp(g_cum[:, :, -1:] - g_cum, max=50.0).exp()
-    // key_gdiff = key * g_diff.unsqueeze(-1)
-    // kgdmulvnew = (key_gdiff).transpose(-1, -2) @ v_new
-    // last_recurrent_state = last_recurrent_state * g_last + kgdmulvnew
-
-    ggml_tensor * g_cum_last =
-        ggml_cont(ctx0, ggml_view_4d(ctx0, g_cumsum_t, g_cumsum_t->ne[0], 1, g_cumsum_t->ne[2], g_cumsum_t->ne[3],
-                                    g_cumsum_t->nb[1], g_cumsum_t->nb[2], g_cumsum_t->nb[3],
-                                    g_cumsum_t->nb[0] * (g_cumsum_t->ne[1] - 1)));
-
-    cb(g_cum_last, "g_cum_last", il);
-
-    ggml_tensor * gexp_last =
-        ggml_reshape_4d(ctx0, ggml_exp(ctx0, g_cum_last), 1, 1, g_cum_last->ne[0] * g_cum_last->ne[2], g_cum_last->ne[3]);
-
-    cb(gexp_last, "gexp_last", il);
-
-    ggml_tensor * g_cum_last_3d =
-        ggml_reshape_3d(ctx0, g_cum_last, g_cum_last->ne[0], g_cum_last->ne[2], g_cum_last->ne[3]);
-
-    cb(g_cum_last_3d, "g_cum_last_3d", il);
-
-    ggml_tensor * g_cumsum_3d = ggml_reshape_3d(ctx0, g_cumsum, g_cumsum->ne[0], g_cumsum->ne[2], g_cumsum->ne[3]);
-
-    cb(g_cumsum_3d, "g_cumsum_3d", il);
-
-    ggml_tensor * g_diff = ggml_neg(ctx0, ggml_sub(ctx0, g_cumsum_3d, g_cum_last_3d));
-
-    cb(g_diff, "g_diff", il);
-
-    ggml_tensor * g_diff_exp = ggml_exp(ctx0, g_diff);
-
-    cb(g_diff_exp, "g_diff_exp", il);
-
-    ggml_tensor * key_gdiff = ggml_mul(ctx0, k,
-                                    ggml_reshape_4d(ctx0, g_diff_exp, 1, g_diff_exp->ne[0], g_diff_exp->ne[1],
-                                                    g_diff_exp->ne[2] * g_diff_exp->ne[3]));
-
-    cb(key_gdiff, "key_gdiff", il);
-
-    ggml_tensor * kgdmulvnew = ggml_mul_mat(ctx0, v_new_t, ggml_cont(ctx0, ggml_transpose(ctx0, key_gdiff)));
-
-    cb(kgdmulvnew, "kgdmulvnew", il);
-
-    state = ggml_add(ctx0, ggml_mul(ctx0, state, gexp_last), kgdmulvnew);
+    // Compute the attention output
+    // core_attn_out = (last_recurrent_state * q_t.unsqueeze(-1)).sum(dim=-2)
+    ggml_tensor * q_t_unsqueezed = ggml_reshape_4d(ctx0, q, 1, S_v, H_v, n_seqs);  // unsqueeze q_t
+    ggml_tensor * state_q        = ggml_mul(ctx0, state, q_t_unsqueezed);
+    // again, since it's over dim = -2, transpose, sum, transpose back
+    ggml_tensor * core_attn_out =
+        ggml_transpose(ctx0, ggml_sum_rows(ctx0, ggml_cont(ctx0, ggml_transpose(ctx0, state_q))));
 
+    // core_attn_out should be [S_v, 1, H_v, n_seqs] after this
+    cb(core_attn_out, "output_tokens", il);
     cb(state, "new_state", il);
 
-    // flatten output
-    ggml_tensor * flat_output =
-        ggml_cont_1d(ctx0, ggml_permute(ctx0, core_attn_out, 0, 2, 1, 3), S_v * H_v * n_tokens * n_seqs);
-
-    ggml_tensor * flat_state = ggml_cont_1d(ctx0, state, S_v * S_v * H_v * n_seqs);
+    // flatten output, no need to permute since n_tokens is 1 so [S_v, 1, H_v, n_seqs] and [S_v, H_v, 1, n_seqs] are equivalent memory-layout wise
+    ggml_tensor * flat_output = ggml_reshape_1d(ctx0, core_attn_out, S_v * H_v * n_tokens * n_seqs);
+    ggml_tensor * flat_state  = ggml_reshape_1d(ctx0, state, S_v * S_v * H_v * n_seqs);
 
     return ggml_concat(ctx0, flat_output, flat_state, 0);
 }
@@ -712,6 +528,7 @@ ggml_tensor * llm_build_qwen3next::build_layer_attn_linear(
         ggml_tensor *        cur,
         ggml_tensor *        causal_mask,
         ggml_tensor *        identity,
+        ggml_tensor *        diag_mask,
         int                  il) {
     const auto * mctx_cur = inp->mctx;
 
@@ -737,11 +554,11 @@ ggml_tensor * llm_build_qwen3next::build_layer_attn_linear(
     cb(mixed_ba, "linear_attn_mixed_ba", il);
 
     int64_t       qkvz_new_dim        = 2 * head_k_dim + 2 * head_v_dim * (num_v_heads / num_k_heads);
-    ggml_tensor * mixed_qkvz_reshaped = ggml_cont_4d(ctx0, mixed_qkvz, qkvz_new_dim, num_k_heads, n_seq_tokens, n_seqs);
+    ggml_tensor * mixed_qkvz_reshaped = ggml_reshape_4d(ctx0, mixed_qkvz, qkvz_new_dim, num_k_heads, n_seq_tokens, n_seqs);
 
     // Reshape mixed_ba: [batch, seq_len, hidden_size] -> [batch, seq_len, num_k_heads, 2*num_v_heads/num_k_heads]
     int64_t       ba_new_dim        = 2 * num_v_heads / num_k_heads;
-    ggml_tensor * mixed_ba_reshaped = ggml_cont_4d(ctx0, mixed_ba, ba_new_dim, num_k_heads, n_seq_tokens, n_seqs);
+    ggml_tensor * mixed_ba_reshaped = ggml_reshape_4d(ctx0, mixed_ba, ba_new_dim, num_k_heads, n_seq_tokens, n_seqs);
 
     // Split mixed_ba into b and a (beta and alpha parameters)
     int64_t split_sizes_ba[2] = {
@@ -762,8 +579,6 @@ ggml_tensor * llm_build_qwen3next::build_layer_attn_linear(
     ggml_tensor * beta  = ggml_cont_3d(ctx0, b, num_v_heads, n_seq_tokens, n_seqs);
     ggml_tensor * alpha = ggml_cont_3d(ctx0, a, num_v_heads, n_seq_tokens, n_seqs);
 
-    GGML_ASSERT(ggml_nelements(beta) + ggml_nelements(alpha) == ggml_nelements(mixed_ba));
-
     ggml_tensor * alpha_biased   = ggml_add(ctx0, alpha, model.layers[il].ssm_dt);
     ggml_tensor * alpha_softplus = ggml_softplus(ctx0, alpha_biased);
     cb(alpha_softplus, "a_softplus", il);
@@ -799,9 +614,6 @@ ggml_tensor * llm_build_qwen3next::build_layer_attn_linear(
                                    (split_sizes_qkvz[0] + split_sizes_qkvz[1] + split_sizes_qkvz[2]) * sizeof(float));
     cb(z, "z", il);
 
-    GGML_ASSERT(ggml_nelements(query) + ggml_nelements(key) + ggml_nelements(value) + ggml_nelements(z) ==
-                ggml_nelements(mixed_qkvz));
-
     // After creating query, key, and value_reshaped, reshape each to flatten the head dimensions
     // query: [head_k_dim, num_k_heads, n_tokens, n_seqs] -> [head_k_dim * num_k_heads, n_tokens, n_seqs]
     ggml_tensor * query_flat = ggml_cont_3d(ctx0, query, head_k_dim * num_k_heads, n_seq_tokens, n_seqs);
@@ -925,10 +737,13 @@ ggml_tensor * llm_build_qwen3next::build_layer_attn_linear(
     cb(k_conv, "k_conv_predelta", il);
     cb(v_conv, "v_conv_predelta", il);
 
-    // Choose between build_delta_net_chunking and build_delta_net_recurrent based on n_tokens
-    ggml_tensor * attn_out = n_seq_tokens > CHUNK_SIZE ?
-        build_delta_net_chunking (q_conv, k_conv, v_conv, gate, beta, state, causal_mask, identity, il) :
-        build_delta_net_recurrent(q_conv, k_conv, v_conv, gate, beta, state, causal_mask, identity, il);
+    // Choose between build_delta_net_chunking, build_delta_net_recurrent, and build_delta_net_autoregressive based on n_tokens
+    ggml_tensor * attn_out;
+    if (n_seq_tokens == 1) {
+        attn_out = build_delta_net_autoregressive(q_conv, k_conv, v_conv, gate, beta, state, il);
+    } else {
+        attn_out = build_delta_net_chunking(q_conv, k_conv, v_conv, gate, beta, state, causal_mask, identity, diag_mask, il);
+    }
     cb(attn_out, "attn_out", il);
 
     // The tensors were concatenated 1d, so we need to extract them 1d as well

From 7b1db3d3b770d0affbf3aadee033e1614280085f Mon Sep 17 00:00:00 2001
From: Xuan-Son Nguyen <son@huggingface.co>
Date: Tue, 16 Dec 2025 12:01:27 +0100
Subject: [PATCH 2/6] arg: clarify auto kvu/np being set on server (#17997)

* arg: clarify auto kvu/np being set on server

* improve docs

* use invalid_argument
---
 common/arg.cpp                  | 50 +++++++++++++++++++++++++--------
 examples/gen-docs/gen-docs.cpp  |  9 +++---
 tools/completion/completion.cpp |  3 --
 tools/mtmd/mtmd-cli.cpp         |  2 --
 tools/server/README.md          | 13 ++++-----
 tools/server/server.cpp         |  9 ++----
 6 files changed, 51 insertions(+), 35 deletions(-)

diff --git a/common/arg.cpp b/common/arg.cpp
index acf4c8f8a8..f2aec895ba 100644
--- a/common/arg.cpp
+++ b/common/arg.cpp
@@ -835,6 +835,19 @@ bool common_arg_utils::is_autoy(const std::string & value) {
 }
 
 common_params_context common_params_parser_init(common_params & params, llama_example ex, void(*print_usage)(int, char **)) {
+    // per-example default params
+    // we define here to make sure it's included in llama-gen-docs
+    if (ex == LLAMA_EXAMPLE_COMPLETION) {
+        params.use_jinja = false;   // disable jinja by default
+
+    } else if (ex == LLAMA_EXAMPLE_MTMD) {
+        params.use_jinja = false;   // disable jinja by default
+        params.sampling.temp = 0.2; // lower temp by default for better quality
+
+    } else if (ex == LLAMA_EXAMPLE_SERVER) {
+        params.n_parallel = -1;     // auto by default
+    }
+
     params.use_color = tty_can_use_colors();
 
     // load dynamic backends
@@ -1107,7 +1120,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
     ).set_env("LLAMA_ARG_SWA_FULL"));
     add_opt(common_arg(
         {"--ctx-checkpoints", "--swa-checkpoints"}, "N",
-        string_format("max number of context checkpoints to create per slot (default: %d)\n"
+        string_format("max number of context checkpoints to create per slot (default: %d)"
             "[(more info)](https://github.com/ggml-org/llama.cpp/pull/15293)", params.n_ctx_checkpoints),
         [](common_params & params, int value) {
             params.n_ctx_checkpoints = value;
@@ -1115,7 +1128,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
     ).set_env("LLAMA_ARG_CTX_CHECKPOINTS").set_examples({LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_CLI}));
     add_opt(common_arg(
         {"--cache-ram", "-cram"}, "N",
-        string_format("set the maximum cache size in MiB (default: %d, -1 - no limit, 0 - disable)\n"
+        string_format("set the maximum cache size in MiB (default: %d, -1 - no limit, 0 - disable)"
             "[(more info)](https://github.com/ggml-org/llama.cpp/pull/16391)", params.cache_ram_mib),
         [](common_params & params, int value) {
             params.cache_ram_mib = value;
@@ -1123,12 +1136,11 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
     ).set_env("LLAMA_ARG_CACHE_RAM").set_examples({LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_CLI}));
     add_opt(common_arg(
         {"--kv-unified", "-kvu"},
-        string_format("use single unified KV buffer for the KV cache of all sequences (default: %s)\n"
-            "[(more info)](https://github.com/ggml-org/llama.cpp/pull/14363)", params.kv_unified ? "true" : "false"),
+        "use single unified KV buffer shared across all sequences (default: enabled if number of slots is auto)",
         [](common_params & params) {
             params.kv_unified = true;
         }
-    ).set_env("LLAMA_ARG_KV_UNIFIED"));
+    ).set_env("LLAMA_ARG_KV_UNIFIED").set_examples({LLAMA_EXAMPLE_SERVER}));
     add_opt(common_arg(
         {"--context-shift"},
         {"--no-context-shift"},
@@ -1888,13 +1900,27 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
             LOG_WRN("DEPRECATED: --defrag-thold is deprecated and no longer necessary to specify\n");
         }
     ).set_env("LLAMA_ARG_DEFRAG_THOLD"));
-    add_opt(common_arg(
-        {"-np", "--parallel"}, "N",
-        string_format("number of parallel sequences to decode (default: %d)", params.n_parallel),
-        [](common_params & params, int value) {
-            params.n_parallel = value;
-        }
-    ).set_env("LLAMA_ARG_N_PARALLEL"));
+    if (ex == LLAMA_EXAMPLE_SERVER) {
+        // this is to make sure this option appears in the server-specific section of the help message
+        add_opt(common_arg(
+            {"-np", "--parallel"}, "N",
+            string_format("number of server slots (default: %d, -1 = auto)", params.n_parallel),
+            [](common_params & params, int value) {
+                if (value == 0) {
+                    throw std::invalid_argument("error: invalid value for n_parallel\n");
+                }
+                params.n_parallel = value;
+            }
+        ).set_env("LLAMA_ARG_N_PARALLEL").set_examples({LLAMA_EXAMPLE_SERVER}));
+    } else {
+        add_opt(common_arg(
+            {"-np", "--parallel"}, "N",
+            string_format("number of parallel sequences to decode (default: %d)", params.n_parallel),
+            [](common_params & params, int value) {
+                params.n_parallel = value;
+            }
+        ).set_env("LLAMA_ARG_N_PARALLEL"));
+    }
     add_opt(common_arg(
         {"-ns", "--sequences"}, "N",
         string_format("number of sequences to decode (default: %d)", params.n_sequences),
diff --git a/examples/gen-docs/gen-docs.cpp b/examples/gen-docs/gen-docs.cpp
index e9f7bf9313..dc76c4cf53 100644
--- a/examples/gen-docs/gen-docs.cpp
+++ b/examples/gen-docs/gen-docs.cpp
@@ -48,7 +48,7 @@ static void write_table(std::ofstream & file, std::vector<common_arg *> & opts)
     }
 }
 
-static void export_md(std::string fname, llama_example ex) {
+static void export_md(std::string fname, llama_example ex, std::string name) {
     std::ofstream file(fname, std::ofstream::out | std::ofstream::trunc);
 
     common_params params;
@@ -72,13 +72,14 @@ static void export_md(std::string fname, llama_example ex) {
     write_table(file, common_options);
     file << "\n\n**Sampling params**\n\n";
     write_table(file, sparam_options);
-    file << "\n\n**Example-specific params**\n\n";
+    file << "\n\n**" << name << "-specific params**\n\n";
     write_table(file, specific_options);
 }
 
 int main(int, char **) {
-    export_md("autogen-main.md", LLAMA_EXAMPLE_COMPLETION);
-    export_md("autogen-server.md", LLAMA_EXAMPLE_SERVER);
+    // TODO: add CLI
+    export_md("autogen-completion.md", LLAMA_EXAMPLE_COMPLETION, "Tool");
+    export_md("autogen-server.md", LLAMA_EXAMPLE_SERVER, "Server");
 
     return 0;
 }
diff --git a/tools/completion/completion.cpp b/tools/completion/completion.cpp
index 85480f3369..29770515f5 100644
--- a/tools/completion/completion.cpp
+++ b/tools/completion/completion.cpp
@@ -87,9 +87,6 @@ int main(int argc, char ** argv) {
     common_params params;
     g_params = &params;
 
-    // disable jinja by default
-    params.use_jinja = false;
-
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_COMPLETION, print_usage)) {
         return 1;
     }
diff --git a/tools/mtmd/mtmd-cli.cpp b/tools/mtmd/mtmd-cli.cpp
index 332d2049e5..3ee1c2eccf 100644
--- a/tools/mtmd/mtmd-cli.cpp
+++ b/tools/mtmd/mtmd-cli.cpp
@@ -270,8 +270,6 @@ int main(int argc, char ** argv) {
     ggml_time_init();
 
     common_params params;
-    params.use_jinja = false;   // disable jinja by default
-    params.sampling.temp = 0.2; // lower temp by default for better quality
 
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_MTMD, show_additional_info)) {
         return 1;
diff --git a/tools/server/README.md b/tools/server/README.md
index ef4990faf1..9a2b9b1f36 100644
--- a/tools/server/README.md
+++ b/tools/server/README.md
@@ -52,7 +52,6 @@ For the ful list of features, please refer to [server's changelog](https://githu
 | `-ub, --ubatch-size N` | physical maximum batch size (default: 512)<br/>(env: LLAMA_ARG_UBATCH) |
 | `--keep N` | number of tokens to keep from the initial prompt (default: 0, -1 = all) |
 | `--swa-full` | use full-size SWA cache (default: false)<br/>[(more info)](https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)<br/>(env: LLAMA_ARG_SWA_FULL) |
-| `--kv-unified, -kvu` | use single unified KV buffer for the KV cache of all sequences (default: false)<br/>[(more info)](https://github.com/ggml-org/llama.cpp/pull/14363)<br/>(env: LLAMA_ARG_KV_UNIFIED) |
 | `-fa, --flash-attn [on\|off\|auto]` | set Flash Attention use ('on', 'off', or 'auto', default: 'auto')<br/>(env: LLAMA_ARG_FLASH_ATTN) |
 | `--perf, --no-perf` | whether to enable internal libllama performance timings (default: false)<br/>(env: LLAMA_ARG_PERF) |
 | `-e, --escape, --no-escape` | whether to process escapes sequences (\n, \r, \t, \', \", \\) (default: true) |
@@ -67,11 +66,10 @@ For the ful list of features, please refer to [server's changelog](https://githu
 | `--yarn-beta-fast N` | YaRN: low correction dim or beta (default: -1.0)<br/>(env: LLAMA_ARG_YARN_BETA_FAST) |
 | `-kvo, --kv-offload, -nkvo, --no-kv-offload` | whether to enable KV cache offloading (default: enabled)<br/>(env: LLAMA_ARG_KV_OFFLOAD) |
 | `--repack, -nr, --no-repack` | whether to enable weight repacking (default: enabled)<br/>(env: LLAMA_ARG_REPACK) |
-| `--no-host` | bypass host buffer allowing extra buffers to be used<br/>(env: LLAMA_ARG_HOST) |
+| `--no-host` | bypass host buffer allowing extra buffers to be used<br/>(env: LLAMA_ARG_NO_HOST) |
 | `-ctk, --cache-type-k TYPE` | KV cache data type for K<br/>allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1<br/>(default: f16)<br/>(env: LLAMA_ARG_CACHE_TYPE_K) |
 | `-ctv, --cache-type-v TYPE` | KV cache data type for V<br/>allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1<br/>(default: f16)<br/>(env: LLAMA_ARG_CACHE_TYPE_V) |
 | `-dt, --defrag-thold N` | KV cache defragmentation threshold (DEPRECATED)<br/>(env: LLAMA_ARG_DEFRAG_THOLD) |
-| `-np, --parallel N` | number of parallel sequences to decode (default: 1)<br/>(env: LLAMA_ARG_N_PARALLEL) |
 | `--mlock` | force system to keep model in RAM rather than swapping or compressing<br/>(env: LLAMA_ARG_MLOCK) |
 | `--mmap, --no-mmap` | whether to memory-map model (if disabled, slower load but may reduce pageouts if not using mlock) (default: enabled)<br/>(env: LLAMA_ARG_MMAP) |
 | `--numa TYPE` | attempt optimizations that help on some NUMA systems<br/>- distribute: spread execution evenly over all nodes<br/>- isolate: only spawn threads on CPUs on the node that execution started on<br/>- numactl: use the CPU map provided by numactl<br/>if run without this previously, it is recommended to drop the system page cache before using this<br/>see https://github.com/ggml-org/llama.cpp/issues/1437<br/>(env: LLAMA_ARG_NUMA) |
@@ -150,19 +148,20 @@ For the ful list of features, please refer to [server's changelog](https://githu
 | `-jf, --json-schema-file FILE` | File containing a JSON schema to constrain generations (https://json-schema.org/), e.g. `{}` for any JSON object<br/>For schemas w/ external $refs, use --grammar + example/json_schema_to_grammar.py instead |
 
 
-**Example-specific params**
+**Server-specific params**
 
 | Argument | Explanation |
 | -------- | ----------- |
-| `--ctx-checkpoints, --swa-checkpoints N` | max number of context checkpoints to create per slot (default: 8)<br/>[(more info)](https://github.com/ggml-org/llama.cpp/pull/15293)<br/>(env: LLAMA_ARG_CTX_CHECKPOINTS) |
-| `--cache-ram, -cram N` | set the maximum cache size in MiB (default: 8192, -1 - no limit, 0 - disable)<br/>[(more info)](https://github.com/ggml-org/llama.cpp/pull/16391)<br/>(env: LLAMA_ARG_CACHE_RAM) |
+| `--ctx-checkpoints, --swa-checkpoints N` | max number of context checkpoints to create per slot (default: 8)[(more info)](https://github.com/ggml-org/llama.cpp/pull/15293)<br/>(env: LLAMA_ARG_CTX_CHECKPOINTS) |
+| `--cache-ram, -cram N` | set the maximum cache size in MiB (default: 8192, -1 - no limit, 0 - disable)[(more info)](https://github.com/ggml-org/llama.cpp/pull/16391)<br/>(env: LLAMA_ARG_CACHE_RAM) |
+| `--kv-unified, -kvu` | use single unified KV buffer shared across all sequences (default: enabled if number of slots is auto)<br/>(env: LLAMA_ARG_KV_UNIFIED) |
 | `--context-shift, --no-context-shift` | whether to use context shift on infinite text generation (default: disabled)<br/>(env: LLAMA_ARG_CONTEXT_SHIFT) |
 | `-r, --reverse-prompt PROMPT` | halt generation at PROMPT, return control in interactive mode<br/> |
 | `-sp, --special` | special tokens output enabled (default: false) |
 | `--warmup, --no-warmup` | whether to perform warmup with an empty run (default: enabled) |
 | `--spm-infill` | use Suffix/Prefix/Middle pattern for infill (instead of Prefix/Suffix/Middle) as some models prefer this. (default: disabled) |
 | `--pooling {none,mean,cls,last,rank}` | pooling type for embeddings, use model default if unspecified<br/>(env: LLAMA_ARG_POOLING) |
-| `-cb, --cont-batching, -nocb, --no-cont-batching` | whether to enable continuous batching (a.k.a dynamic batching) (default: enabled)<br/>(env: LLAMA_ARG_CONT_BATCHING) |
+| `-np, --parallel N` | number of server slots (default: -1, -1 = auto)<br/>(env: LLAMA_ARG_N_PARALLEL) |
 | `-cb, --cont-batching, -nocb, --no-cont-batching` | whether to enable continuous batching (a.k.a dynamic batching) (default: enabled)<br/>(env: LLAMA_ARG_CONT_BATCHING) |
 | `-mm, --mmproj FILE` | path to a multimodal projector file. see tools/mtmd/README.md<br/>note: if -hf is used, this argument can be omitted<br/>(env: LLAMA_ARG_MMPROJ) |
 | `-mmu, --mmproj-url URL` | URL to a multimodal projector file. see tools/mtmd/README.md<br/>(env: LLAMA_ARG_MMPROJ_URL) |
diff --git a/tools/server/server.cpp b/tools/server/server.cpp
index d5bef3df44..235ae4e8c0 100644
--- a/tools/server/server.cpp
+++ b/tools/server/server.cpp
@@ -73,13 +73,8 @@ int main(int argc, char ** argv, char ** envp) {
         return 1;
     }
 
-    // TODO: should we have a separate n_parallel parameter for the server?
-    //       https://github.com/ggml-org/llama.cpp/pull/16736#discussion_r2483763177
-    // TODO: this is a common configuration that is suitable for most local use cases
-    //       however, overriding the parameters is a bit confusing - figure out something more intuitive
-    if (params.n_parallel == 1 && params.kv_unified == false && !params.has_speculative()) {
-        LOG_WRN("%s: setting n_parallel = 4 and kv_unified = true (add -kvu to disable this)\n", __func__);
-
+    if (params.n_parallel < 0) {
+        LOG_INF("%s: n_parallel is set to auto, using n_parallel = 4 and kv_unified = true\n", __func__);
         params.n_parallel = 4;
         params.kv_unified = true;
     }

From 7f2b2f3c778f430edc57d5728641317e9ac5a505 Mon Sep 17 00:00:00 2001
From: Xuan-Son Nguyen <son@huggingface.co>
Date: Tue, 16 Dec 2025 13:22:30 +0100
Subject: [PATCH 3/6] arch: refactor LLM_TENSOR_NAMES (#18051)

* arch: refactor LLM_TENSOR_NAMES

* update docs

* typo

* fix LLM_ARCH_NEMOTRON_H_MOE

* show more meaningful error message on missing tensor

* fix and tested LLM_ARCH_NEMOTRON_H_MOE
---
 docs/development/HOWTO-add-model.md |    2 +-
 src/llama-arch.cpp                  | 4169 ++++++++++++---------------
 src/llama-arch.h                    |   10 +-
 3 files changed, 1897 insertions(+), 2284 deletions(-)

diff --git a/docs/development/HOWTO-add-model.md b/docs/development/HOWTO-add-model.md
index 9d1452e3f0..b6870f6e25 100644
--- a/docs/development/HOWTO-add-model.md
+++ b/docs/development/HOWTO-add-model.md
@@ -97,7 +97,7 @@ The model params and tensors layout must be defined in `llama.cpp` source files:
 1. Define a new `llm_arch` enum value in `src/llama-arch.h`.
 2. In `src/llama-arch.cpp`:
     - Add the architecture name to the `LLM_ARCH_NAMES` map.
-    - Add the tensor mappings to the `LLM_TENSOR_NAMES` map.
+    - Add the list of model tensors to `llm_get_tensor_names` (you may also need to update `LLM_TENSOR_NAMES`)
 3. Add any non-standard metadata loading in the `llama_model_loader` constructor in `src/llama-model-loader.cpp`.
 4. If the model has a RoPE operation, add a case for the architecture in `llama_model_rope_type` function in `src/llama-model.cpp`.
 
diff --git a/src/llama-arch.cpp b/src/llama-arch.cpp
index 05b12a6072..8caf80afcf 100644
--- a/src/llama-arch.cpp
+++ b/src/llama-arch.cpp
@@ -3,6 +3,7 @@
 #include "llama-impl.h"
 
 #include <map>
+#include <set>
 
 static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
     { LLM_ARCH_CLIP,             "clip"             }, // dummy, only used by llama-quantize
@@ -302,2286 +303,1884 @@ static const std::map<llm_kv, const char *> LLM_KV_NAMES = {
     { LLM_KV_TOKENIZER_MIDDLE_ID, "tokenizer.ggml.middle_token_id" },
 };
 
-static const std::map<llm_arch, std::map<llm_tensor, const char *>> LLM_TENSOR_NAMES = {
-    {
-        LLM_ARCH_CLIP,
-        {},
-    },
-    {
-        LLM_ARCH_LLAMA,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd" },
-            { LLM_TENSOR_FFN_GATE_INP,    "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_GATE_EXP,    "blk.%d.ffn_gate.%d" },
-            { LLM_TENSOR_FFN_DOWN_EXP,    "blk.%d.ffn_down.%d" },
-            { LLM_TENSOR_FFN_UP_EXP,      "blk.%d.ffn_up.%d" },
-            { LLM_TENSOR_FFN_GATE_EXPS,   "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,   "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,     "blk.%d.ffn_up_exps" },
-        },
-    },
-    {
-        LLM_ARCH_ARCEE,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_AFMOE,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_POST_NORM,  "blk.%d.post_attention_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_Q_NORM,     "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_ATTN_K_NORM,     "blk.%d.attn_k_norm" },
-            { LLM_TENSOR_ATTN_GATE,       "blk.%d.attn_gate" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_POST_NORM,   "blk.%d.post_ffw_norm" },
-            { LLM_TENSOR_FFN_GATE_INP,    "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_GATE_EXPS,   "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,   "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,     "blk.%d.ffn_up_exps" },
-            { LLM_TENSOR_FFN_GATE_SHEXP,  "blk.%d.ffn_gate_shexp" },
-            { LLM_TENSOR_FFN_UP_SHEXP,    "blk.%d.ffn_up_shexp" },
-            { LLM_TENSOR_FFN_DOWN_SHEXP,  "blk.%d.ffn_down_shexp" },
-            { LLM_TENSOR_FFN_EXP_PROBS_B, "blk.%d.exp_probs_b" },
-        },
-    },
-    {
-        LLM_ARCH_LLAMA4,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd" },
-            { LLM_TENSOR_FFN_GATE_INP,    "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_GATE_EXP,    "blk.%d.ffn_gate.%d" },
-            { LLM_TENSOR_FFN_DOWN_EXP,    "blk.%d.ffn_down.%d" },
-            { LLM_TENSOR_FFN_UP_EXP,      "blk.%d.ffn_up.%d" },
-            { LLM_TENSOR_FFN_GATE_EXPS,   "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,   "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,     "blk.%d.ffn_up_exps" },
-            { LLM_TENSOR_FFN_GATE_SHEXP,  "blk.%d.ffn_gate_shexp" },
-            { LLM_TENSOR_FFN_DOWN_SHEXP,  "blk.%d.ffn_down_shexp" },
-            { LLM_TENSOR_FFN_UP_SHEXP,    "blk.%d.ffn_up_shexp" },
-        },
-    },
-    {
-        LLM_ARCH_DECI,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd" },
-            { LLM_TENSOR_FFN_GATE_INP,    "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_GATE_EXP,    "blk.%d.ffn_gate.%d" },
-            { LLM_TENSOR_FFN_DOWN_EXP,    "blk.%d.ffn_down.%d" },
-            { LLM_TENSOR_FFN_UP_EXP,      "blk.%d.ffn_up.%d" },
-            { LLM_TENSOR_FFN_GATE_EXPS,   "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,   "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,     "blk.%d.ffn_up_exps" },
-        },
-    },
-    {
-        LLM_ARCH_BAICHUAN,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_FALCON,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_NORM_2,     "blk.%d.attn_norm_2" },
-            { LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_GROK,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd" },
-            { LLM_TENSOR_FFN_GATE_INP,    "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_GATE_EXP,    "blk.%d.ffn_gate.%d" },
-            { LLM_TENSOR_FFN_DOWN_EXP,    "blk.%d.ffn_down.%d" },
-            { LLM_TENSOR_FFN_UP_EXP,      "blk.%d.ffn_up.%d" },
-            { LLM_TENSOR_FFN_GATE_EXPS,   "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,   "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,     "blk.%d.ffn_up_exps" },
-            { LLM_TENSOR_FFN_POST_NORM,   "blk.%d.post_ffw_norm" },
-            { LLM_TENSOR_LAYER_OUT_NORM,  "blk.%d.layer_output_norm" },
-            { LLM_TENSOR_ATTN_OUT_NORM,   "blk.%d.attn_output_norm" },
-        },
-    },
-    {
-        LLM_ARCH_GPT2,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_POS_EMBD,        "position_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-        },
-    },
-    {
-        LLM_ARCH_GPTJ,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-        },
-    },
-    {
-        LLM_ARCH_GPTNEOX,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_MPT,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output"},
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_ACT,         "blk.%d.ffn.act" },
-            { LLM_TENSOR_POS_EMBD,        "position_embd" },
-            { LLM_TENSOR_ATTN_Q_NORM,     "blk.%d.attn_q_norm"},
-            { LLM_TENSOR_ATTN_K_NORM,     "blk.%d.attn_k_norm"},
-        },
-    },
-    {
-        LLM_ARCH_STARCODER,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_POS_EMBD,        "position_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-        },
-    },
-    {
-        LLM_ARCH_REFACT,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_BERT,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_TOKEN_EMBD_NORM, "token_embd_norm" },
-            { LLM_TENSOR_TOKEN_TYPES,     "token_types" },
-            { LLM_TENSOR_POS_EMBD,        "position_embd" },
-            { LLM_TENSOR_ATTN_OUT_NORM,   "blk.%d.attn_output_norm" },
-            { LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_LAYER_OUT_NORM,  "blk.%d.layer_output_norm" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_CLS,             "cls" },
-            { LLM_TENSOR_CLS_OUT,         "cls.output" },
-        },
-    },
-    {
-        LLM_ARCH_NOMIC_BERT,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_TOKEN_EMBD_NORM, "token_embd_norm" },
-            { LLM_TENSOR_TOKEN_TYPES,     "token_types" },
-            { LLM_TENSOR_ATTN_OUT_NORM,   "blk.%d.attn_output_norm" },
-            { LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_LAYER_OUT_NORM,  "blk.%d.layer_output_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_NOMIC_BERT_MOE,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_TOKEN_EMBD_NORM, "token_embd_norm" },
-            { LLM_TENSOR_TOKEN_TYPES,     "token_types" },
-            { LLM_TENSOR_ATTN_OUT_NORM,   "blk.%d.attn_output_norm" },
-            { LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_LAYER_OUT_NORM,  "blk.%d.layer_output_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_GATE_INP,    "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,   "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,     "blk.%d.ffn_up_exps" },
-        },
-    },
-    {
-        LLM_ARCH_NEO_BERT,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_ENC_OUTPUT_NORM, "enc.output_norm" },
-            { LLM_TENSOR_CLS,             "cls" },
-            { LLM_TENSOR_CLS_OUT,         "cls.output" },
-        },
-    },
-    {
-        LLM_ARCH_JINA_BERT_V2,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_TOKEN_EMBD_NORM, "token_embd_norm" },
-            { LLM_TENSOR_TOKEN_TYPES,     "token_types" },
-            { LLM_TENSOR_ATTN_NORM_2,     "blk.%d.attn_norm_2" },
-            { LLM_TENSOR_ATTN_OUT_NORM,   "blk.%d.attn_output_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_Q_NORM,     "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_K_NORM,     "blk.%d.attn_k_norm" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_LAYER_OUT_NORM,  "blk.%d.layer_output_norm" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_CLS,             "cls" },
-        },
-    },
-    {
-        LLM_ARCH_JINA_BERT_V3,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_TOKEN_EMBD_NORM, "token_embd_norm" },
-            { LLM_TENSOR_TOKEN_TYPES,     "token_types" },
-            { LLM_TENSOR_ATTN_OUT_NORM,   "blk.%d.attn_output_norm" },
-            { LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_LAYER_OUT_NORM,  "blk.%d.layer_output_norm" },
-        },
-    },
-    {
-        LLM_ARCH_BLOOM,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_TOKEN_EMBD_NORM, "token_embd_norm" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-        },
-    },
-    {
-        LLM_ARCH_STABLELM,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_ATTN_Q_NORM,     "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_ATTN_K_NORM,     "blk.%d.attn_k_norm" },
-        },
-    },
-    {
-        LLM_ARCH_QWEN,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_QWEN2,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_QWEN2VL,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_QWEN2MOE,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,         "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,        "output_norm" },
-            { LLM_TENSOR_OUTPUT,             "output" },
-            { LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,             "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,             "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,             "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,           "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE_INP,       "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_GATE_EXPS,      "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,      "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,        "blk.%d.ffn_up_exps" },
-            { LLM_TENSOR_FFN_GATE_INP_SHEXP, "blk.%d.ffn_gate_inp_shexp" },
-            { LLM_TENSOR_FFN_GATE_SHEXP,     "blk.%d.ffn_gate_shexp" },
-            { LLM_TENSOR_FFN_DOWN_SHEXP,     "blk.%d.ffn_down_shexp" },
-            { LLM_TENSOR_FFN_UP_SHEXP,       "blk.%d.ffn_up_shexp" },
-        },
-    },
-    {
-        LLM_ARCH_QWEN3,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_CLS_OUT,         "cls.output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_Q_NORM,     "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_K_NORM,     "blk.%d.attn_k_norm" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_QWEN3MOE,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,         "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,        "output_norm" },
-            { LLM_TENSOR_OUTPUT,             "output" },
-            { LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,             "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_Q_NORM,        "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_ATTN_K,             "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_K_NORM,        "blk.%d.attn_k_norm" },
-            { LLM_TENSOR_ATTN_V,             "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,           "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE_INP,       "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_GATE_EXPS,      "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,      "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,        "blk.%d.ffn_up_exps" },
-        },
-    },
-    {
-        LLM_ARCH_QWEN3NEXT,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,         "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,        "output_norm" },
-            { LLM_TENSOR_OUTPUT,             "output" },
-            { LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_POST_NORM,     "blk.%d.post_attention_norm" },
-            { LLM_TENSOR_ATTN_Q,             "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_Q_NORM,        "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_ATTN_K,             "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_K_NORM,        "blk.%d.attn_k_norm" },
-            { LLM_TENSOR_ATTN_V,             "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,           "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE_INP,       "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_GATE_EXPS,      "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,      "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,        "blk.%d.ffn_up_exps" },
-            { LLM_TENSOR_FFN_GATE_INP_SHEXP, "blk.%d.ffn_gate_inp_shexp" },
-            { LLM_TENSOR_FFN_GATE_SHEXP,     "blk.%d.ffn_gate_shexp" },
-            { LLM_TENSOR_FFN_DOWN_SHEXP,     "blk.%d.ffn_down_shexp" },
-            { LLM_TENSOR_FFN_UP_SHEXP,       "blk.%d.ffn_up_shexp" },
-            { LLM_TENSOR_SSM_A_NOSCAN,       "blk.%d.ssm_a" },
-            { LLM_TENSOR_SSM_CONV1D,         "blk.%d.ssm_conv1d" },
-            { LLM_TENSOR_SSM_DT,             "blk.%d.ssm_dt" },
-            { LLM_TENSOR_SSM_BETA_ALPHA,     "blk.%d.ssm_ba" },
-            { LLM_TENSOR_SSM_IN,             "blk.%d.ssm_in" },
-            { LLM_TENSOR_SSM_NORM,           "blk.%d.ssm_norm" },
-            { LLM_TENSOR_SSM_OUT,            "blk.%d.ssm_out" },
-        },
-    },
-    {
-        LLM_ARCH_QWEN3VL,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_Q_NORM,     "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_K_NORM,     "blk.%d.attn_k_norm" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_QWEN3VLMOE,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,         "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,        "output_norm" },
-            { LLM_TENSOR_OUTPUT,             "output" },
-            { LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,             "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_Q_NORM,        "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_ATTN_K,             "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_K_NORM,        "blk.%d.attn_k_norm" },
-            { LLM_TENSOR_ATTN_V,             "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,           "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE_INP,       "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_GATE_EXPS,      "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,      "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,        "blk.%d.ffn_up_exps" },
-        },
-    },
-    {
-        LLM_ARCH_PHI2,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_PHI3,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,         "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,        "output_norm" },
-            { LLM_TENSOR_OUTPUT,             "output" },
-            { LLM_TENSOR_ROPE_FACTORS_LONG,  "rope_factors_long" },
-            { LLM_TENSOR_ROPE_FACTORS_SHORT, "rope_factors_short" },
-            { LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_QKV,           "blk.%d.attn_qkv" },
-            { LLM_TENSOR_ATTN_Q,             "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,             "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,             "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,           "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_DOWN,           "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,             "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_PHIMOE,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,         "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,        "output_norm" },
-            { LLM_TENSOR_OUTPUT,             "output" },
-            { LLM_TENSOR_ROPE_FACTORS_LONG,  "rope_factors_long" },
-            { LLM_TENSOR_ROPE_FACTORS_SHORT, "rope_factors_short" },
-            { LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_QKV,           "blk.%d.attn_qkv" },
-            { LLM_TENSOR_ATTN_Q,             "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,             "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,             "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,           "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE_INP,       "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_GATE_EXPS,      "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,      "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,        "blk.%d.ffn_up_exps" },
-        },
-    },
-    {
-        LLM_ARCH_PLAMO,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_PLAMO2,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },
-            { LLM_TENSOR_ATTN_Q_NORM,     "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_ATTN_K_NORM,     "blk.%d.attn_k_norm" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_SSM_IN,          "blk.%d.ssm_in" },
-            { LLM_TENSOR_SSM_CONV1D,      "blk.%d.ssm_conv1d" },
-            { LLM_TENSOR_SSM_X,           "blk.%d.ssm_x" },
-            { LLM_TENSOR_SSM_DT,          "blk.%d.ssm_dt" },
-            { LLM_TENSOR_SSM_A,           "blk.%d.ssm_a" },
-            { LLM_TENSOR_SSM_D,           "blk.%d.ssm_d" },
-            { LLM_TENSOR_SSM_OUT,         "blk.%d.ssm_out" },
-            { LLM_TENSOR_SSM_DT_NORM,     "blk.%d.ssm_dt_norm" },
-            { LLM_TENSOR_SSM_B_NORM,      "blk.%d.ssm_b_norm" },
-            { LLM_TENSOR_SSM_C_NORM,      "blk.%d.ssm_c_norm" },
-            { LLM_TENSOR_ATTN_POST_NORM,  "blk.%d.post_attention_norm" },
-            { LLM_TENSOR_FFN_POST_NORM,   "blk.%d.post_ffw_norm" },
-        },
-    },
-    {
-        LLM_ARCH_CODESHELL,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_ORION,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_INTERNLM2,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_MINICPM,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },
-            { LLM_TENSOR_ROPE_FACTORS_LONG,  "rope_factors_long" },
-            { LLM_TENSOR_ROPE_FACTORS_SHORT, "rope_factors_short" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd" },
-            { LLM_TENSOR_FFN_GATE_INP,    "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_GATE_EXP,    "blk.%d.ffn_gate.%d" },
-            { LLM_TENSOR_FFN_DOWN_EXP,    "blk.%d.ffn_down.%d" },
-            { LLM_TENSOR_FFN_UP_EXP,      "blk.%d.ffn_up.%d" },
-        },
-    },
-    {
-        LLM_ARCH_MINICPM3,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,         "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,        "output_norm" },
-            { LLM_TENSOR_OUTPUT,             "output" },
-            { LLM_TENSOR_ROPE_FACTORS_LONG,  "rope_factors_long" },
-            { LLM_TENSOR_ROPE_FACTORS_SHORT, "rope_factors_short" },
-            { LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q_A_NORM,      "blk.%d.attn_q_a_norm" },
-            { LLM_TENSOR_ATTN_KV_A_NORM,     "blk.%d.attn_kv_a_norm" },
-            { LLM_TENSOR_ATTN_Q,             "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_Q_A,           "blk.%d.attn_q_a" },
-            { LLM_TENSOR_ATTN_Q_B,           "blk.%d.attn_q_b" },
-            { LLM_TENSOR_ATTN_KV_A_MQA,      "blk.%d.attn_kv_a_mqa" },
-            { LLM_TENSOR_ATTN_KV_B,          "blk.%d.attn_kv_b" },
-            { LLM_TENSOR_ATTN_OUT,           "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,           "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_UP,             "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_DOWN,           "blk.%d.ffn_down" },
-        },
-    },
-    {
-        LLM_ARCH_GEMMA,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_GEMMA2,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_POST_NORM,  "blk.%d.post_attention_norm" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_POST_NORM,   "blk.%d.post_ffw_norm" },
-        },
-    },
-    {
-        LLM_ARCH_GEMMA3,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_Q_NORM,     "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_K_NORM,     "blk.%d.attn_k_norm" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_POST_NORM,  "blk.%d.post_attention_norm" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_POST_NORM,   "blk.%d.post_ffw_norm" },
-        },
-    },
-    {
-        LLM_ARCH_GEMMA3N,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,           "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,          "output_norm" },
-            { LLM_TENSOR_ATTN_NORM,            "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,               "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_Q_NORM,          "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_ATTN_K,               "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_K_NORM,          "blk.%d.attn_k_norm" },
-            { LLM_TENSOR_ATTN_V,               "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,             "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_POST_NORM,       "blk.%d.post_attention_norm" },
-            { LLM_TENSOR_FFN_NORM,             "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,             "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,             "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,               "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_POST_NORM,        "blk.%d.post_ffw_norm" },
-            { LLM_TENSOR_PER_LAYER_TOKEN_EMBD, "per_layer_token_embd" },
-            { LLM_TENSOR_PER_LAYER_MODEL_PROJ, "per_layer_model_proj" },
-            { LLM_TENSOR_PER_LAYER_PROJ_NORM,  "per_layer_proj_norm" },
-            { LLM_TENSOR_ALTUP_UNEMBD_PROJ,    "altup_unembd_proj" },
-            { LLM_TENSOR_ALTUP_PROJ,           "altup_proj" },
-            { LLM_TENSOR_PER_LAYER_INP_GATE,   "blk.%d.inp_gate" },
-            { LLM_TENSOR_PER_LAYER_PROJ,       "blk.%d.proj" },
-            { LLM_TENSOR_PER_LAYER_POST_NORM,  "blk.%d.post_norm" },
-            { LLM_TENSOR_ALTUP_CORRECT_COEF,   "blk.%d.altup_correct_coef" },
-            { LLM_TENSOR_ALTUP_CORRECT_SCALE,  "blk.%d.altup_correct_scale" },
-            { LLM_TENSOR_ALTUP_PREDICT_COEF,   "blk.%d.altup_predict_coef" },
-            { LLM_TENSOR_ALTUP_ROUTER,         "blk.%d.altup_router" },
-            { LLM_TENSOR_ALTUP_ROUTER_NORM,    "blk.%d.altup_router_norm" },
-            { LLM_TENSOR_LAUREL_L,             "blk.%d.laurel_l" },
-            { LLM_TENSOR_LAUREL_R,             "blk.%d.laurel_r" },
-            { LLM_TENSOR_LAUREL_POST_NORM,     "blk.%d.laurel_post_norm" },
-        },
-    },
-    {
-        LLM_ARCH_GEMMA_EMBEDDING,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_DENSE_2_OUT,     "dense_2" },
-            { LLM_TENSOR_DENSE_3_OUT,     "dense_3" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_Q_NORM,     "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_K_NORM,     "blk.%d.attn_k_norm" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_POST_NORM,  "blk.%d.post_attention_norm" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_POST_NORM,   "blk.%d.post_ffw_norm" },
-        },
-    },
-    {
-        LLM_ARCH_STARCODER2,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_MAMBA,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_SSM_IN,          "blk.%d.ssm_in" },
-            { LLM_TENSOR_SSM_CONV1D,      "blk.%d.ssm_conv1d" },
-            { LLM_TENSOR_SSM_X,           "blk.%d.ssm_x" },
-            { LLM_TENSOR_SSM_DT,          "blk.%d.ssm_dt" },
-            { LLM_TENSOR_SSM_A,           "blk.%d.ssm_a" },
-            { LLM_TENSOR_SSM_D,           "blk.%d.ssm_d" },
-            { LLM_TENSOR_SSM_OUT,         "blk.%d.ssm_out" },
-        },
-    },
-    {
-        LLM_ARCH_MAMBA2,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_SSM_IN,          "blk.%d.ssm_in" },
-            { LLM_TENSOR_SSM_CONV1D,      "blk.%d.ssm_conv1d" },
-            { LLM_TENSOR_SSM_DT,          "blk.%d.ssm_dt" },
-            { LLM_TENSOR_SSM_A,           "blk.%d.ssm_a" },
-            { LLM_TENSOR_SSM_D,           "blk.%d.ssm_d" },
-            { LLM_TENSOR_SSM_NORM,        "blk.%d.ssm_norm" },
-            { LLM_TENSOR_SSM_OUT,         "blk.%d.ssm_out" },
-        },
-    },
-    {
-        LLM_ARCH_JAMBA,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_SSM_IN,          "blk.%d.ssm_in" },
-            { LLM_TENSOR_SSM_CONV1D,      "blk.%d.ssm_conv1d" },
-            { LLM_TENSOR_SSM_X,           "blk.%d.ssm_x" },
-            { LLM_TENSOR_SSM_DT,          "blk.%d.ssm_dt" },
-            { LLM_TENSOR_SSM_DT_NORM,     "blk.%d.ssm_dt_norm" },
-            { LLM_TENSOR_SSM_A,           "blk.%d.ssm_a" },
-            { LLM_TENSOR_SSM_B_NORM,      "blk.%d.ssm_b_norm" },
-            { LLM_TENSOR_SSM_C_NORM,      "blk.%d.ssm_c_norm" },
-            { LLM_TENSOR_SSM_D,           "blk.%d.ssm_d" },
-            { LLM_TENSOR_SSM_OUT,         "blk.%d.ssm_out" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_GATE_INP,    "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_GATE_EXPS,   "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,   "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,     "blk.%d.ffn_up_exps" },
-        },
-    },
-    {
-        LLM_ARCH_FALCON_H1,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_SSM_IN,          "blk.%d.ssm_in" },
-            { LLM_TENSOR_SSM_CONV1D,      "blk.%d.ssm_conv1d" },
-            { LLM_TENSOR_SSM_DT,          "blk.%d.ssm_dt" },
-            { LLM_TENSOR_SSM_A,           "blk.%d.ssm_a" },
-            { LLM_TENSOR_SSM_D,           "blk.%d.ssm_d" },
-            { LLM_TENSOR_SSM_NORM,        "blk.%d.ssm_norm" },
-            { LLM_TENSOR_SSM_OUT,         "blk.%d.ssm_out" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_XVERSE,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_COMMAND_R,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_ATTN_Q_NORM,     "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_ATTN_K_NORM,     "blk.%d.attn_k_norm" },
-        },
-    },
-    {
-        LLM_ARCH_COHERE2,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_DBRX,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_OUT_NORM,   "blk.%d.attn_output_norm" },
-            { LLM_TENSOR_FFN_GATE_INP,    "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_GATE_EXPS,   "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,   "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,     "blk.%d.ffn_up_exps" },
-        },
-    },
-    {
-        LLM_ARCH_OLMO,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_OLMO2,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_POST_NORM,  "blk.%d.post_attention_norm" },
-            { LLM_TENSOR_ATTN_Q_NORM,     "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_ATTN_K_NORM,     "blk.%d.attn_k_norm" },
-            { LLM_TENSOR_FFN_POST_NORM,   "blk.%d.post_ffw_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_OLMOE,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,         "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,        "output_norm" },
-            { LLM_TENSOR_OUTPUT,             "output" },
-            { LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,             "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,             "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,             "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,           "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_Q_NORM,        "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_ATTN_K_NORM,        "blk.%d.attn_k_norm" },
-            { LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE_INP,       "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_GATE_EXPS,      "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,      "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,        "blk.%d.ffn_up_exps" },
-        },
-    },
-    {
-        LLM_ARCH_OPENELM,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },
-            { LLM_TENSOR_ATTN_Q_NORM,     "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_ATTN_K_NORM,     "blk.%d.attn_k_norm" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_ARCTIC,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_GATE_INP,    "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_NORM_EXPS,   "blk.%d.ffn_norm_exps" },
-            { LLM_TENSOR_FFN_GATE_EXPS,   "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,   "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,     "blk.%d.ffn_up_exps" },
-        },
-    },
-    {
-        LLM_ARCH_DEEPSEEK,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,         "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,        "output_norm" },
-            { LLM_TENSOR_OUTPUT,             "output" },
-            { LLM_TENSOR_ROPE_FREQS,         "rope_freqs" },
-            { LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,             "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,             "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,             "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,           "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_ROT_EMBD,      "blk.%d.attn_rot_embd" },
-            { LLM_TENSOR_FFN_GATE_INP,       "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,           "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,           "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,             "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_GATE_EXPS,      "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,      "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,        "blk.%d.ffn_up_exps" },
-            { LLM_TENSOR_FFN_GATE_INP_SHEXP, "blk.%d.ffn_gate_inp_shexp" },
-            { LLM_TENSOR_FFN_GATE_SHEXP,     "blk.%d.ffn_gate_shexp" },
-            { LLM_TENSOR_FFN_DOWN_SHEXP,     "blk.%d.ffn_down_shexp" },
-            { LLM_TENSOR_FFN_UP_SHEXP,       "blk.%d.ffn_up_shexp" },
-        },
-    },
-    {
-        LLM_ARCH_DEEPSEEK2,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,         "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,        "output_norm" },
-            { LLM_TENSOR_OUTPUT,             "output" },
-            { LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q_A_NORM,      "blk.%d.attn_q_a_norm" },
-            { LLM_TENSOR_ATTN_KV_A_NORM,     "blk.%d.attn_kv_a_norm" },
-            { LLM_TENSOR_ATTN_Q,             "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_Q_A,           "blk.%d.attn_q_a" },
-            { LLM_TENSOR_ATTN_Q_B,           "blk.%d.attn_q_b" },
-            { LLM_TENSOR_ATTN_KV_A_MQA,      "blk.%d.attn_kv_a_mqa" },
-            { LLM_TENSOR_ATTN_KV_B,          "blk.%d.attn_kv_b" },
-            { LLM_TENSOR_ATTN_K_B,           "blk.%d.attn_k_b" },
-            { LLM_TENSOR_ATTN_V_B,           "blk.%d.attn_v_b" },
-            { LLM_TENSOR_ATTN_OUT,           "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,           "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_UP,             "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_DOWN,           "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_GATE_INP,       "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_GATE_EXPS,      "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,      "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,        "blk.%d.ffn_up_exps" },
-            { LLM_TENSOR_FFN_GATE_INP_SHEXP, "blk.%d.ffn_gate_inp_shexp" },
-            { LLM_TENSOR_FFN_GATE_SHEXP,     "blk.%d.ffn_gate_shexp" },
-            { LLM_TENSOR_FFN_DOWN_SHEXP,     "blk.%d.ffn_down_shexp" },
-            { LLM_TENSOR_FFN_UP_SHEXP,       "blk.%d.ffn_up_shexp" },
-            { LLM_TENSOR_FFN_EXP_PROBS_B,    "blk.%d.exp_probs_b" },
-        },
-    },
-    {
-        LLM_ARCH_PLM,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,         "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,        "output_norm" },
-            { LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,             "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_KV_A_MQA,      "blk.%d.attn_kv_a_mqa" },
-            { LLM_TENSOR_ATTN_KV_A_NORM,     "blk.%d.attn_kv_a_norm" },
-            { LLM_TENSOR_ATTN_KV_B,          "blk.%d.attn_kv_b" },
-            { LLM_TENSOR_ATTN_OUT,           "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_DOWN,           "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,             "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_CHATGLM,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-        },
-    },
-    {
-        LLM_ARCH_GLM4,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_ATTN_POST_NORM,  "blk.%d.post_attention_norm" },
-            { LLM_TENSOR_FFN_POST_NORM,   "blk.%d.post_ffw_norm" },
-        },
-    },
-    {
-        LLM_ARCH_GLM4_MOE,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,         "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,        "output_norm" },
-            { LLM_TENSOR_OUTPUT,             "output" },
-            { LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_POST_NORM,     "blk.%d.post_attention_norm" },
-            { LLM_TENSOR_ATTN_Q,             "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,             "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,             "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,           "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_Q_NORM,        "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_ATTN_K_NORM,        "blk.%d.attn_k_norm" },
-            { LLM_TENSOR_FFN_GATE,           "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,           "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,             "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_GATE_INP,       "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_GATE_EXPS,      "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,      "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,        "blk.%d.ffn_up_exps" },
-            { LLM_TENSOR_FFN_GATE_SHEXP,     "blk.%d.ffn_gate_shexp" },
-            { LLM_TENSOR_FFN_DOWN_SHEXP,     "blk.%d.ffn_down_shexp" },
-            { LLM_TENSOR_FFN_UP_SHEXP,       "blk.%d.ffn_up_shexp" },
-            { LLM_TENSOR_FFN_EXP_PROBS_B,    "blk.%d.exp_probs_b" },
-            // NextN/MTP tensors - preserved but unused (in final layer, dynamic layer number)
-            { LLM_TENSOR_NEXTN_EH_PROJ,      "blk.%d.nextn.eh_proj" },
-            { LLM_TENSOR_NEXTN_EMBED_TOKENS, "blk.%d.nextn.embed_tokens" },
-            { LLM_TENSOR_NEXTN_ENORM,        "blk.%d.nextn.enorm" },
-            { LLM_TENSOR_NEXTN_HNORM,        "blk.%d.nextn.hnorm" },
-            { LLM_TENSOR_NEXTN_SHARED_HEAD_HEAD, "blk.%d.nextn.shared_head_head" },
-            { LLM_TENSOR_NEXTN_SHARED_HEAD_NORM, "blk.%d.nextn.shared_head_norm" },
-        },
-    },
-    {
-        LLM_ARCH_BITNET,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,         "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,        "output_norm" },
-            { LLM_TENSOR_ATTN_Q,             "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,             "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,             "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,           "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_SUB_NORM,      "blk.%d.attn_sub_norm" },
-            { LLM_TENSOR_FFN_GATE,           "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,           "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,             "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_SUB_NORM,       "blk.%d.ffn_sub_norm" },
-        },
-    },
-    {
-        LLM_ARCH_T5,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,           "token_embd" },
-            { LLM_TENSOR_OUTPUT,               "output" },
-            { LLM_TENSOR_DEC_OUTPUT_NORM,      "dec.output_norm" },
-            { LLM_TENSOR_DEC_ATTN_NORM,        "dec.blk.%d.attn_norm" },
-            { LLM_TENSOR_DEC_ATTN_Q,           "dec.blk.%d.attn_q" },
-            { LLM_TENSOR_DEC_ATTN_K,           "dec.blk.%d.attn_k" },
-            { LLM_TENSOR_DEC_ATTN_V,           "dec.blk.%d.attn_v" },
-            { LLM_TENSOR_DEC_ATTN_OUT,         "dec.blk.%d.attn_o" },
-            { LLM_TENSOR_DEC_ATTN_REL_B,       "dec.blk.%d.attn_rel_b" },
-            { LLM_TENSOR_DEC_CROSS_ATTN_NORM,  "dec.blk.%d.cross_attn_norm" },
-            { LLM_TENSOR_DEC_CROSS_ATTN_Q,     "dec.blk.%d.cross_attn_q" },
-            { LLM_TENSOR_DEC_CROSS_ATTN_K,     "dec.blk.%d.cross_attn_k" },
-            { LLM_TENSOR_DEC_CROSS_ATTN_V,     "dec.blk.%d.cross_attn_v" },
-            { LLM_TENSOR_DEC_CROSS_ATTN_OUT,   "dec.blk.%d.cross_attn_o" },
-            { LLM_TENSOR_DEC_CROSS_ATTN_REL_B, "dec.blk.%d.cross_attn_rel_b" },
-            { LLM_TENSOR_DEC_FFN_NORM,         "dec.blk.%d.ffn_norm" },
-            { LLM_TENSOR_DEC_FFN_GATE,         "dec.blk.%d.ffn_gate" },
-            { LLM_TENSOR_DEC_FFN_DOWN,         "dec.blk.%d.ffn_down" },
-            { LLM_TENSOR_DEC_FFN_UP,           "dec.blk.%d.ffn_up" },
-            { LLM_TENSOR_ENC_OUTPUT_NORM,      "enc.output_norm" },
-            { LLM_TENSOR_ENC_ATTN_NORM,        "enc.blk.%d.attn_norm" },
-            { LLM_TENSOR_ENC_ATTN_Q,           "enc.blk.%d.attn_q" },
-            { LLM_TENSOR_ENC_ATTN_K,           "enc.blk.%d.attn_k" },
-            { LLM_TENSOR_ENC_ATTN_V,           "enc.blk.%d.attn_v" },
-            { LLM_TENSOR_ENC_ATTN_OUT,         "enc.blk.%d.attn_o" },
-            { LLM_TENSOR_ENC_ATTN_REL_B,       "enc.blk.%d.attn_rel_b" },
-            { LLM_TENSOR_ENC_FFN_NORM,         "enc.blk.%d.ffn_norm" },
-            { LLM_TENSOR_ENC_FFN_GATE,         "enc.blk.%d.ffn_gate" },
-            { LLM_TENSOR_ENC_FFN_DOWN,         "enc.blk.%d.ffn_down" },
-            { LLM_TENSOR_ENC_FFN_UP,           "enc.blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_T5ENCODER,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,           "token_embd" },
-            { LLM_TENSOR_OUTPUT,               "output" },
-            { LLM_TENSOR_ENC_OUTPUT_NORM,      "enc.output_norm" },
-            { LLM_TENSOR_ENC_ATTN_NORM,        "enc.blk.%d.attn_norm" },
-            { LLM_TENSOR_ENC_ATTN_Q,           "enc.blk.%d.attn_q" },
-            { LLM_TENSOR_ENC_ATTN_K,           "enc.blk.%d.attn_k" },
-            { LLM_TENSOR_ENC_ATTN_V,           "enc.blk.%d.attn_v" },
-            { LLM_TENSOR_ENC_ATTN_OUT,         "enc.blk.%d.attn_o" },
-            { LLM_TENSOR_ENC_ATTN_REL_B,       "enc.blk.%d.attn_rel_b" },
-            { LLM_TENSOR_ENC_FFN_NORM,         "enc.blk.%d.ffn_norm" },
-            { LLM_TENSOR_ENC_FFN_GATE,         "enc.blk.%d.ffn_gate" },
-            { LLM_TENSOR_ENC_FFN_DOWN,         "enc.blk.%d.ffn_down" },
-            { LLM_TENSOR_ENC_FFN_UP,           "enc.blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_JAIS,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-        },
-    },
-    {
-        LLM_ARCH_NEMOTRON,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_NEMOTRON_H,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,     "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,    "output_norm" },
-            { LLM_TENSOR_OUTPUT,         "output" },
-            { LLM_TENSOR_ATTN_NORM,      "blk.%d.attn_norm" },
-            // mamba(2) ssm layers
-            { LLM_TENSOR_SSM_IN,         "blk.%d.ssm_in" },
-            { LLM_TENSOR_SSM_CONV1D,     "blk.%d.ssm_conv1d" },
-            { LLM_TENSOR_SSM_DT,         "blk.%d.ssm_dt" },
-            { LLM_TENSOR_SSM_A,          "blk.%d.ssm_a" },
-            { LLM_TENSOR_SSM_D,          "blk.%d.ssm_d" },
-            { LLM_TENSOR_SSM_NORM,       "blk.%d.ssm_norm" },
-            { LLM_TENSOR_SSM_OUT,        "blk.%d.ssm_out" },
-            // attention layers
-            { LLM_TENSOR_ATTN_Q,         "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,         "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,         "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,       "blk.%d.attn_output" },
-            // dense FFN
-            { LLM_TENSOR_FFN_DOWN,       "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,         "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_NEMOTRON_H_MOE,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,     "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,    "output_norm" },
-            { LLM_TENSOR_OUTPUT,         "output" },
-            { LLM_TENSOR_ATTN_NORM,      "blk.%d.attn_norm" },
-            // mamba(2) ssm layers
-            { LLM_TENSOR_SSM_IN,         "blk.%d.ssm_in" },
-            { LLM_TENSOR_SSM_CONV1D,     "blk.%d.ssm_conv1d" },
-            { LLM_TENSOR_SSM_DT,         "blk.%d.ssm_dt" },
-            { LLM_TENSOR_SSM_A,          "blk.%d.ssm_a" },
-            { LLM_TENSOR_SSM_D,          "blk.%d.ssm_d" },
-            { LLM_TENSOR_SSM_NORM,       "blk.%d.ssm_norm" },
-            { LLM_TENSOR_SSM_OUT,        "blk.%d.ssm_out" },
-            // attention layers
-            { LLM_TENSOR_ATTN_Q,         "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,         "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,         "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,       "blk.%d.attn_output" },
-            // dense FFN
-            { LLM_TENSOR_FFN_DOWN,       "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,         "blk.%d.ffn_up" },
-            // MoE FFN (for MoE layers)
-            { LLM_TENSOR_FFN_GATE_INP,   "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_UP_EXPS,    "blk.%d.ffn_up_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,  "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_EXP_PROBS_B,"blk.%d.exp_probs_b" },
-            // MoE shared expert layer
-            { LLM_TENSOR_FFN_DOWN_SHEXP,  "blk.%d.ffn_down_shexp" },
-            { LLM_TENSOR_FFN_UP_SHEXP,    "blk.%d.ffn_up_shexp" },
-        },
-    },
-    {
-        LLM_ARCH_EXAONE,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_EXAONE4,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_Q_NORM,     "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_K_NORM,     "blk.%d.attn_k_norm" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_POST_NORM,  "blk.%d.post_attention_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_POST_NORM,   "blk.%d.post_ffw_norm" },
-        }
-    },
-    {
-        LLM_ARCH_RWKV6,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,                "token_embd" },
-            { LLM_TENSOR_TOKEN_EMBD_NORM,           "token_embd_norm" },
-            { LLM_TENSOR_OUTPUT_NORM,               "output_norm" },
-            { LLM_TENSOR_OUTPUT,                    "output" },
-            { LLM_TENSOR_ATTN_NORM,                 "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_NORM_2,               "blk.%d.attn_norm_2" },
-            { LLM_TENSOR_TIME_MIX_W1,               "blk.%d.time_mix_w1" },
-            { LLM_TENSOR_TIME_MIX_W2,               "blk.%d.time_mix_w2" },
-            { LLM_TENSOR_TIME_MIX_LERP_X,           "blk.%d.time_mix_lerp_x" },
-            { LLM_TENSOR_TIME_MIX_LERP_W,           "blk.%d.time_mix_lerp_w" },
-            { LLM_TENSOR_TIME_MIX_LERP_K,           "blk.%d.time_mix_lerp_k" },
-            { LLM_TENSOR_TIME_MIX_LERP_V,           "blk.%d.time_mix_lerp_v" },
-            { LLM_TENSOR_TIME_MIX_LERP_R,           "blk.%d.time_mix_lerp_r" },
-            { LLM_TENSOR_TIME_MIX_LERP_G,           "blk.%d.time_mix_lerp_g" },
-            { LLM_TENSOR_TIME_MIX_LERP_FUSED,       "blk.%d.time_mix_lerp_fused" },
-            { LLM_TENSOR_TIME_MIX_FIRST,            "blk.%d.time_mix_first" },
-            { LLM_TENSOR_TIME_MIX_DECAY,            "blk.%d.time_mix_decay" },
-            { LLM_TENSOR_TIME_MIX_DECAY_W1,         "blk.%d.time_mix_decay_w1" },
-            { LLM_TENSOR_TIME_MIX_DECAY_W2,         "blk.%d.time_mix_decay_w2" },
-            { LLM_TENSOR_TIME_MIX_KEY,              "blk.%d.time_mix_key" },
-            { LLM_TENSOR_TIME_MIX_VALUE,            "blk.%d.time_mix_value" },
-            { LLM_TENSOR_TIME_MIX_RECEPTANCE,       "blk.%d.time_mix_receptance" },
-            { LLM_TENSOR_TIME_MIX_GATE,             "blk.%d.time_mix_gate" },
-            { LLM_TENSOR_TIME_MIX_LN,               "blk.%d.time_mix_ln" },
-            { LLM_TENSOR_TIME_MIX_OUTPUT,           "blk.%d.time_mix_output" },
-            { LLM_TENSOR_CHANNEL_MIX_LERP_K,        "blk.%d.channel_mix_lerp_k" },
-            { LLM_TENSOR_CHANNEL_MIX_LERP_R,        "blk.%d.channel_mix_lerp_r" },
-            { LLM_TENSOR_CHANNEL_MIX_KEY,           "blk.%d.channel_mix_key" },
-            { LLM_TENSOR_CHANNEL_MIX_VALUE,         "blk.%d.channel_mix_value" },
-            { LLM_TENSOR_CHANNEL_MIX_RECEPTANCE,    "blk.%d.channel_mix_receptance" },
-        },
-    },
-    {
-        LLM_ARCH_RWKV6QWEN2,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,                "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,               "output_norm" },
-            { LLM_TENSOR_OUTPUT,                    "output" },
-            { LLM_TENSOR_ATTN_NORM,                 "blk.%d.attn_norm" },
-            { LLM_TENSOR_TIME_MIX_W1,               "blk.%d.time_mix_w1" },
-            { LLM_TENSOR_TIME_MIX_W2,               "blk.%d.time_mix_w2" },
-            { LLM_TENSOR_TIME_MIX_LERP_X,           "blk.%d.time_mix_lerp_x" },
-            { LLM_TENSOR_TIME_MIX_LERP_FUSED,       "blk.%d.time_mix_lerp_fused" },
-            { LLM_TENSOR_TIME_MIX_FIRST,            "blk.%d.time_mix_first" },
-            { LLM_TENSOR_TIME_MIX_DECAY,            "blk.%d.time_mix_decay" },
-            { LLM_TENSOR_TIME_MIX_DECAY_W1,         "blk.%d.time_mix_decay_w1" },
-            { LLM_TENSOR_TIME_MIX_DECAY_W2,         "blk.%d.time_mix_decay_w2" },
-            { LLM_TENSOR_TIME_MIX_KEY,              "blk.%d.time_mix_key" },
-            { LLM_TENSOR_TIME_MIX_VALUE,            "blk.%d.time_mix_value" },
-            { LLM_TENSOR_TIME_MIX_RECEPTANCE,       "blk.%d.time_mix_receptance" },
-            { LLM_TENSOR_TIME_MIX_GATE,             "blk.%d.time_mix_gate" },
-            { LLM_TENSOR_TIME_MIX_OUTPUT,           "blk.%d.time_mix_output" },
-            { LLM_TENSOR_FFN_NORM,                  "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,                  "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,                  "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,                    "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_RWKV7,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,                "token_embd" },
-            { LLM_TENSOR_TOKEN_EMBD_NORM,           "token_embd_norm" },
-            { LLM_TENSOR_OUTPUT_NORM,               "output_norm" },
-            { LLM_TENSOR_OUTPUT,                    "output" },
-            { LLM_TENSOR_ATTN_NORM,                 "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_NORM_2,               "blk.%d.attn_norm_2" },
-            { LLM_TENSOR_TIME_MIX_W0,               "blk.%d.time_mix_w0" },
-            { LLM_TENSOR_TIME_MIX_W1,               "blk.%d.time_mix_w1" },
-            { LLM_TENSOR_TIME_MIX_W2,               "blk.%d.time_mix_w2" },
-            { LLM_TENSOR_TIME_MIX_A0,               "blk.%d.time_mix_a0" },
-            { LLM_TENSOR_TIME_MIX_A1,               "blk.%d.time_mix_a1" },
-            { LLM_TENSOR_TIME_MIX_A2,               "blk.%d.time_mix_a2" },
-            { LLM_TENSOR_TIME_MIX_V0,               "blk.%d.time_mix_v0" },
-            { LLM_TENSOR_TIME_MIX_V1,               "blk.%d.time_mix_v1" },
-            { LLM_TENSOR_TIME_MIX_V2,               "blk.%d.time_mix_v2" },
-            { LLM_TENSOR_TIME_MIX_G1,               "blk.%d.time_mix_g1" },
-            { LLM_TENSOR_TIME_MIX_G2,               "blk.%d.time_mix_g2" },
-            { LLM_TENSOR_TIME_MIX_K_K,              "blk.%d.time_mix_k_k" },
-            { LLM_TENSOR_TIME_MIX_K_A,              "blk.%d.time_mix_k_a" },
-            { LLM_TENSOR_TIME_MIX_R_K,              "blk.%d.time_mix_r_k" },
-            { LLM_TENSOR_TIME_MIX_LERP_FUSED,       "blk.%d.time_mix_lerp_fused" },
-            { LLM_TENSOR_TIME_MIX_KEY,              "blk.%d.time_mix_key" },
-            { LLM_TENSOR_TIME_MIX_VALUE,            "blk.%d.time_mix_value" },
-            { LLM_TENSOR_TIME_MIX_RECEPTANCE,       "blk.%d.time_mix_receptance" },
-            { LLM_TENSOR_TIME_MIX_LN,               "blk.%d.time_mix_ln" },
-            { LLM_TENSOR_TIME_MIX_OUTPUT,           "blk.%d.time_mix_output" },
-            { LLM_TENSOR_CHANNEL_MIX_LERP_K,        "blk.%d.channel_mix_lerp_k" },
-            { LLM_TENSOR_CHANNEL_MIX_KEY,           "blk.%d.channel_mix_key" },
-            { LLM_TENSOR_CHANNEL_MIX_VALUE,         "blk.%d.channel_mix_value" },
-        },
-    },
-    {
-        LLM_ARCH_ARWKV7,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,                "token_embd" },
-            { LLM_TENSOR_TOKEN_EMBD_NORM,           "token_embd_norm" },
-            { LLM_TENSOR_OUTPUT_NORM,               "output_norm" },
-            { LLM_TENSOR_OUTPUT,                    "output" },
-            { LLM_TENSOR_ATTN_NORM,                 "blk.%d.attn_norm" },
-            { LLM_TENSOR_TIME_MIX_W0,               "blk.%d.time_mix_w0" },
-            { LLM_TENSOR_TIME_MIX_W1,               "blk.%d.time_mix_w1" },
-            { LLM_TENSOR_TIME_MIX_W2,               "blk.%d.time_mix_w2" },
-            { LLM_TENSOR_TIME_MIX_A0,               "blk.%d.time_mix_a0" },
-            { LLM_TENSOR_TIME_MIX_A1,               "blk.%d.time_mix_a1" },
-            { LLM_TENSOR_TIME_MIX_A2,               "blk.%d.time_mix_a2" },
-            { LLM_TENSOR_TIME_MIX_V0,               "blk.%d.time_mix_v0" },
-            { LLM_TENSOR_TIME_MIX_V1,               "blk.%d.time_mix_v1" },
-            { LLM_TENSOR_TIME_MIX_V2,               "blk.%d.time_mix_v2" },
-            { LLM_TENSOR_TIME_MIX_G1,               "blk.%d.time_mix_g1" },
-            { LLM_TENSOR_TIME_MIX_G2,               "blk.%d.time_mix_g2" },
-            { LLM_TENSOR_TIME_MIX_K_K,              "blk.%d.time_mix_k_k" },
-            { LLM_TENSOR_TIME_MIX_K_A,              "blk.%d.time_mix_k_a" },
-            { LLM_TENSOR_TIME_MIX_R_K,              "blk.%d.time_mix_r_k" },
-            { LLM_TENSOR_TIME_MIX_LERP_FUSED,       "blk.%d.time_mix_lerp_fused" },
-            { LLM_TENSOR_TIME_MIX_KEY,              "blk.%d.time_mix_key" },
-            { LLM_TENSOR_TIME_MIX_VALUE,            "blk.%d.time_mix_value" },
-            { LLM_TENSOR_TIME_MIX_RECEPTANCE,       "blk.%d.time_mix_receptance" },
-            { LLM_TENSOR_TIME_MIX_LN,               "blk.%d.time_mix_ln" },
-            { LLM_TENSOR_TIME_MIX_OUTPUT,           "blk.%d.time_mix_output" },
-            { LLM_TENSOR_FFN_NORM,                  "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,                  "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,                  "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,                    "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_GRANITE,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_GRANITE_MOE,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE_INP,    "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_GATE_EXPS,   "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,   "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,     "blk.%d.ffn_up_exps" },
-            { LLM_TENSOR_FFN_GATE_SHEXP,  "blk.%d.ffn_gate_shexp" },
-            { LLM_TENSOR_FFN_DOWN_SHEXP,  "blk.%d.ffn_down_shexp" },
-            { LLM_TENSOR_FFN_UP_SHEXP,    "blk.%d.ffn_up_shexp" },
-        },
-    },
-    {
-        LLM_ARCH_GRANITE_HYBRID,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,     "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,    "output_norm" },
-            { LLM_TENSOR_OUTPUT,         "output" },
-            { LLM_TENSOR_ATTN_NORM,      "blk.%d.attn_norm" },
-            // mamba(2) ssm layers
-            { LLM_TENSOR_SSM_IN,         "blk.%d.ssm_in" },
-            { LLM_TENSOR_SSM_CONV1D,     "blk.%d.ssm_conv1d" },
-            { LLM_TENSOR_SSM_DT,         "blk.%d.ssm_dt" },
-            { LLM_TENSOR_SSM_A,          "blk.%d.ssm_a" },
-            { LLM_TENSOR_SSM_D,          "blk.%d.ssm_d" },
-            { LLM_TENSOR_SSM_NORM,       "blk.%d.ssm_norm" },
-            { LLM_TENSOR_SSM_OUT,        "blk.%d.ssm_out" },
-            // attention layers
-            { LLM_TENSOR_ATTN_Q,         "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,         "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,         "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,       "blk.%d.attn_output" },
-            // dense FFN
-            { LLM_TENSOR_FFN_NORM,       "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,       "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,       "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,         "blk.%d.ffn_up" },
-            // moe FFN
-            { LLM_TENSOR_FFN_NORM,       "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE_INP,   "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_GATE_EXPS,  "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,  "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,    "blk.%d.ffn_up_exps" },
-            // shared expert
-            { LLM_TENSOR_FFN_GATE_SHEXP, "blk.%d.ffn_gate_shexp" },
-            { LLM_TENSOR_FFN_DOWN_SHEXP, "blk.%d.ffn_down_shexp" },
-            { LLM_TENSOR_FFN_UP_SHEXP,   "blk.%d.ffn_up_shexp" },
-        },
-    },
-    {
-        LLM_ARCH_CHAMELEON,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_ATTN_Q_NORM,     "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_ATTN_K_NORM,     "blk.%d.attn_k_norm" },
-        },
-    },
-    {
-        LLM_ARCH_WAVTOKENIZER_DEC,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,        "token_embd" },
-            { LLM_TENSOR_TOKEN_EMBD_NORM,   "token_embd_norm" },
-            { LLM_TENSOR_CONV1D,            "conv1d" },
-            { LLM_TENSOR_CONVNEXT_DW,       "convnext.%d.dw" },
-            { LLM_TENSOR_CONVNEXT_NORM,     "convnext.%d.norm" },
-            { LLM_TENSOR_CONVNEXT_PW1,      "convnext.%d.pw1" },
-            { LLM_TENSOR_CONVNEXT_PW2,      "convnext.%d.pw2" },
-            { LLM_TENSOR_CONVNEXT_GAMMA,    "convnext.%d.gamma" },
-            { LLM_TENSOR_OUTPUT_NORM,       "output_norm" },
-            { LLM_TENSOR_OUTPUT,            "output" },
-            { LLM_TENSOR_POS_NET_CONV1,     "posnet.%d.conv1" },
-            { LLM_TENSOR_POS_NET_CONV2,     "posnet.%d.conv2" },
-            { LLM_TENSOR_POS_NET_NORM,      "posnet.%d.norm" },
-            { LLM_TENSOR_POS_NET_NORM1,     "posnet.%d.norm1" },
-            { LLM_TENSOR_POS_NET_NORM2,     "posnet.%d.norm2" },
-            { LLM_TENSOR_POS_NET_ATTN_NORM, "posnet.%d.attn_norm" },
-            { LLM_TENSOR_POS_NET_ATTN_Q,    "posnet.%d.attn_q" },
-            { LLM_TENSOR_POS_NET_ATTN_K,    "posnet.%d.attn_k" },
-            { LLM_TENSOR_POS_NET_ATTN_V,    "posnet.%d.attn_v" },
-            { LLM_TENSOR_POS_NET_ATTN_OUT,  "posnet.%d.attn_output" },
-        },
-    },
-    {
-        LLM_ARCH_BAILINGMOE,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,         "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,        "output_norm" },
-            { LLM_TENSOR_OUTPUT,             "output" },
-            { LLM_TENSOR_ROPE_FREQS,         "rope_freqs" },
-            { LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,             "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,             "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,             "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,           "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_GATE_INP,       "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE_EXPS,      "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,      "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,        "blk.%d.ffn_up_exps" },
-            { LLM_TENSOR_FFN_GATE_INP_SHEXP, "blk.%d.ffn_gate_inp_shexp" },
-            { LLM_TENSOR_FFN_GATE_SHEXP,     "blk.%d.ffn_gate_shexp" },
-            { LLM_TENSOR_FFN_DOWN_SHEXP,     "blk.%d.ffn_down_shexp" },
-            { LLM_TENSOR_FFN_UP_SHEXP,       "blk.%d.ffn_up_shexp" },
-        },
-    },
-    {
-        LLM_ARCH_BAILINGMOE2,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,         "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,        "output_norm" },
-            { LLM_TENSOR_OUTPUT,             "output" },
-            { LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q_NORM,        "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_ATTN_K_NORM,        "blk.%d.attn_k_norm" },
-            { LLM_TENSOR_ATTN_QKV,           "blk.%d.attn_qkv" },
-            { LLM_TENSOR_ATTN_OUT,           "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_GATE_INP,       "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_EXP_PROBS_B,    "blk.%d.exp_probs_b" },
-            { LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,           "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,           "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,             "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_GATE_EXPS,      "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,      "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,        "blk.%d.ffn_up_exps" },
-            { LLM_TENSOR_FFN_GATE_SHEXP,     "blk.%d.ffn_gate_shexp" },
-            { LLM_TENSOR_FFN_DOWN_SHEXP,     "blk.%d.ffn_down_shexp" },
-            { LLM_TENSOR_FFN_UP_SHEXP,       "blk.%d.ffn_up_shexp" },
-            { LLM_TENSOR_NEXTN_EH_PROJ,      "blk.%d.nextn.eh_proj" },
-            { LLM_TENSOR_NEXTN_EMBED_TOKENS, "blk.%d.nextn.embed_tokens" },
-            { LLM_TENSOR_NEXTN_ENORM,        "blk.%d.nextn.enorm" },
-            { LLM_TENSOR_NEXTN_HNORM,        "blk.%d.nextn.hnorm" },
-            { LLM_TENSOR_NEXTN_SHARED_HEAD_HEAD, "blk.%d.nextn.shared_head_head" },
-            { LLM_TENSOR_NEXTN_SHARED_HEAD_NORM, "blk.%d.nextn.shared_head_norm" },
-            { LLM_TENSOR_LAYER_OUT_NORM,     "blk.%d.layer_output_norm" },
-        },
-    },
-    {
-        LLM_ARCH_DOTS1,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,         "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,        "output_norm" },
-            { LLM_TENSOR_OUTPUT,             "output" },
-            { LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,             "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_Q_NORM,        "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_ATTN_K,             "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_K_NORM,        "blk.%d.attn_k_norm" },
-            { LLM_TENSOR_ATTN_V,             "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,           "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,           "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_UP,             "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_DOWN,           "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_GATE_INP,       "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_GATE_EXPS,      "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,      "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,        "blk.%d.ffn_up_exps" },
-            { LLM_TENSOR_FFN_GATE_INP_SHEXP, "blk.%d.ffn_gate_inp_shexp" },
-            { LLM_TENSOR_FFN_GATE_SHEXP,     "blk.%d.ffn_gate_shexp" },
-            { LLM_TENSOR_FFN_DOWN_SHEXP,     "blk.%d.ffn_down_shexp" },
-            { LLM_TENSOR_FFN_UP_SHEXP,       "blk.%d.ffn_up_shexp" },
-            { LLM_TENSOR_FFN_EXP_PROBS_B,    "blk.%d.exp_probs_b" },
-        }
-    },
-    {
-        LLM_ARCH_ERNIE4_5,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,         "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,        "output_norm" },
-            { LLM_TENSOR_OUTPUT,             "output" },
-            { LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,             "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,             "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,             "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,           "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,           "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,           "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,             "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_ERNIE4_5_MOE,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,         "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,        "output_norm" },
-            { LLM_TENSOR_OUTPUT,             "output" },
-            { LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,             "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,             "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,             "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,           "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,           "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,           "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,             "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_GATE_INP,       "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_GATE_SHEXP,     "blk.%d.ffn_gate_shexp" },
-            { LLM_TENSOR_FFN_DOWN_SHEXP,     "blk.%d.ffn_down_shexp" },
-            { LLM_TENSOR_FFN_UP_SHEXP,       "blk.%d.ffn_up_shexp" },
-            { LLM_TENSOR_FFN_GATE_EXPS,      "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,      "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,        "blk.%d.ffn_up_exps" },
-            { LLM_TENSOR_FFN_EXP_PROBS_B,    "blk.%d.exp_probs_b" },
-        },
-    },
-    {
-        LLM_ARCH_HUNYUAN_MOE,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_Q_NORM,     "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_K_NORM,     "blk.%d.attn_k_norm" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_GATE_INP,    "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE_SHEXP,  "blk.%d.ffn_gate_shexp" },
-            { LLM_TENSOR_FFN_DOWN_SHEXP,  "blk.%d.ffn_down_shexp" },
-            { LLM_TENSOR_FFN_UP_SHEXP,    "blk.%d.ffn_up_shexp" },
-            { LLM_TENSOR_FFN_GATE_EXPS,   "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,   "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,     "blk.%d.ffn_up_exps" },
-        },
-    },
-    {
-        LLM_ARCH_HUNYUAN_DENSE,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_Q_NORM,     "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_K_NORM,     "blk.%d.attn_k_norm" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-
-        },
-    },
-    {
-        LLM_ARCH_SMOLLM3,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,     "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,    "output_norm" },
-            { LLM_TENSOR_OUTPUT,         "output" },
-            { LLM_TENSOR_ATTN_NORM,      "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,         "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,         "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,         "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,       "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,       "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,       "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,       "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,         "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_OPENAI_MOE,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,         "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,        "output_norm" },
-            { LLM_TENSOR_OUTPUT,             "output" },
-            { LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_POST_NORM,     "blk.%d.post_attention_norm" },
-            { LLM_TENSOR_ATTN_Q,             "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,             "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,             "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,           "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_SINKS,         "blk.%d.attn_sinks" },
-            { LLM_TENSOR_FFN_GATE_INP,       "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_GATE_EXPS,      "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,      "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,        "blk.%d.ffn_up_exps" },
-        },
-    },
-    {
-        LLM_ARCH_LFM2,
-        {
-            { LLM_TENSOR_ATTN_NORM,         "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,            "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,            "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,            "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,          "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_K_NORM,       "blk.%d.attn_k_norm" },
-            { LLM_TENSOR_ATTN_Q_NORM,       "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_FFN_DOWN,          "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_GATE,          "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_NORM,          "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_UP,            "blk.%d.ffn_up" },
-            { LLM_TENSOR_SHORTCONV_CONV,    "blk.%d.shortconv.conv" },
-            { LLM_TENSOR_SHORTCONV_INPROJ,  "blk.%d.shortconv.in_proj" },
-            { LLM_TENSOR_SHORTCONV_OUTPROJ, "blk.%d.shortconv.out_proj" },
-            { LLM_TENSOR_TOKEN_EMBD,        "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,       "token_embd_norm" }, // note: wrong tensor name
-            { LLM_TENSOR_OUTPUT,            "output" },
-        }
-    },
-    {
-        LLM_ARCH_LFM2MOE,
-        {
-            { LLM_TENSOR_ATTN_NORM,         "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,            "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,            "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,            "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,          "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_K_NORM,       "blk.%d.attn_k_norm" },
-            { LLM_TENSOR_ATTN_Q_NORM,       "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_FFN_DOWN,          "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_GATE,          "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_NORM,          "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_UP,            "blk.%d.ffn_up" },
-            { LLM_TENSOR_SHORTCONV_CONV,    "blk.%d.shortconv.conv" },
-            { LLM_TENSOR_SHORTCONV_INPROJ,  "blk.%d.shortconv.in_proj" },
-            { LLM_TENSOR_SHORTCONV_OUTPROJ, "blk.%d.shortconv.out_proj" },
-            { LLM_TENSOR_TOKEN_EMBD,        "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,       "token_embd_norm" }, // note: wrong tensor name
-            { LLM_TENSOR_FFN_GATE_INP,      "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_GATE_EXPS,     "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,     "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,       "blk.%d.ffn_up_exps" },
-            { LLM_TENSOR_FFN_EXP_PROBS_B,   "blk.%d.exp_probs_b" },
-        }
-    },
-    {
-        LLM_ARCH_SMALLTHINKER,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,         "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,        "output_norm" },
-            { LLM_TENSOR_OUTPUT,             "output" },
-            { LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,             "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,             "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,             "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,           "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,           "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,           "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,             "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_GATE_INP,       "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_GATE_EXPS,      "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,      "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,        "blk.%d.ffn_up_exps" }
-        },
-    },
-    {
-        LLM_ARCH_APERTUS,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_Q_NORM,     "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_ATTN_K_NORM,     "blk.%d.attn_k_norm" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_DREAM,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_LLADA,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_LLADA_MOE,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,         "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,        "output_norm" },
-            { LLM_TENSOR_OUTPUT,             "output" },
-            { LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,             "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_Q_NORM,        "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_ATTN_K,             "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_K_NORM,        "blk.%d.attn_k_norm" },
-            { LLM_TENSOR_ATTN_V,             "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,           "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE_INP,       "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_GATE_EXPS,      "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,      "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,        "blk.%d.ffn_up_exps" },
-        },
-    },
-    {
-        LLM_ARCH_SEED_OSS,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_POST_NORM,  "blk.%d.post_attention_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_GROVEMOE,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,         "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,        "output_norm" },
-            { LLM_TENSOR_OUTPUT,             "output" },
-            { LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,             "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_Q_NORM,        "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_ATTN_K,             "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_K_NORM,        "blk.%d.attn_k_norm" },
-            { LLM_TENSOR_ATTN_V,             "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,           "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE_INP,       "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_GATE_EXPS,      "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,      "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,        "blk.%d.ffn_up_exps" },
-            { LLM_TENSOR_FFN_GATE_CHEXPS,    "blk.%d.ffn_gate_chexps" },
-            { LLM_TENSOR_FFN_DOWN_CHEXPS,    "blk.%d.ffn_down_chexps" },
-            { LLM_TENSOR_FFN_UP_CHEXPS,      "blk.%d.ffn_up_chexps" },
-        },
-    },
-    {
-        LLM_ARCH_MINIMAX_M2,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,         "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,        "output_norm" },
-            { LLM_TENSOR_OUTPUT,             "output" },
-            { LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,             "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,             "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,             "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,           "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_Q_NORM,        "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_ATTN_K_NORM,        "blk.%d.attn_k_norm" },
-            { LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE_INP,       "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_GATE_EXPS,      "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,      "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,        "blk.%d.ffn_up_exps" },
-            { LLM_TENSOR_FFN_EXP_PROBS_B,    "blk.%d.exp_probs_b" },
-        },
-    },
-    {
-        LLM_ARCH_PANGU_EMBED,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-        },
-    },
-    {
-        LLM_ARCH_COGVLM,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_VISEXP_ATTN_QKV, "blk.%d.vis_attn_qkv" },
-            { LLM_TENSOR_VISEXP_ATTN_OUT, "blk.%d.vis_attn_output" },
-            { LLM_TENSOR_VISEXP_FFN_GATE, "blk.%d.vis_gate" },
-            { LLM_TENSOR_VISEXP_FFN_DOWN, "blk.%d.vis_down" },
-            { LLM_TENSOR_VISEXP_FFN_UP,   "blk.%d.vis_up" },
-        },
-    },
-    {
-        LLM_ARCH_RND1,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,         "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,        "output_norm" },
-            { LLM_TENSOR_OUTPUT,             "output" },
-            { LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,             "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_Q_NORM,        "blk.%d.attn_q_norm" },
-            { LLM_TENSOR_ATTN_K,             "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_K_NORM,        "blk.%d.attn_k_norm" },
-            { LLM_TENSOR_ATTN_V,             "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,           "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE_INP,       "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_GATE_EXPS,      "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,      "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,        "blk.%d.ffn_up_exps" },
-        },
-    },
-    {
-        LLM_ARCH_MISTRAL3,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd" },
-            { LLM_TENSOR_FFN_GATE_INP,    "blk.%d.ffn_gate_inp" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_GATE_EXP,    "blk.%d.ffn_gate.%d" },
-            { LLM_TENSOR_FFN_DOWN_EXP,    "blk.%d.ffn_down.%d" },
-            { LLM_TENSOR_FFN_UP_EXP,      "blk.%d.ffn_up.%d" },
-            { LLM_TENSOR_FFN_GATE_EXPS,   "blk.%d.ffn_gate_exps" },
-            { LLM_TENSOR_FFN_DOWN_EXPS,   "blk.%d.ffn_down_exps" },
-            { LLM_TENSOR_FFN_UP_EXPS,     "blk.%d.ffn_up_exps" },
-        },
-    },
-    {
-        LLM_ARCH_UNKNOWN,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-        },
-    },
+static const std::map<llm_tensor, const char *> LLM_TENSOR_NAMES = {
+    { LLM_TENSOR_TOKEN_EMBD,                             "token_embd" },
+    { LLM_TENSOR_OUTPUT_NORM,                            "output_norm" },
+    { LLM_TENSOR_OUTPUT_NORM_LFM2,                       "token_embd_norm" }, // fix for wrong tensor name
+    { LLM_TENSOR_OUTPUT,                                 "output" },
+    { LLM_TENSOR_ROPE_FREQS,                             "rope_freqs" },
+    { LLM_TENSOR_ATTN_NORM,                              "blk.%d.attn_norm" },
+    { LLM_TENSOR_ATTN_Q,                                 "blk.%d.attn_q" },
+    { LLM_TENSOR_ATTN_K,                                 "blk.%d.attn_k" },
+    { LLM_TENSOR_ATTN_V,                                 "blk.%d.attn_v" },
+    { LLM_TENSOR_ATTN_OUT,                               "blk.%d.attn_output" },
+    { LLM_TENSOR_ATTN_ROT_EMBD,                          "blk.%d.attn_rot_embd" },
+    { LLM_TENSOR_FFN_GATE_INP,                           "blk.%d.ffn_gate_inp" },
+    { LLM_TENSOR_FFN_NORM,                               "blk.%d.ffn_norm" },
+    { LLM_TENSOR_FFN_GATE,                               "blk.%d.ffn_gate" },
+    { LLM_TENSOR_FFN_DOWN,                               "blk.%d.ffn_down" },
+    { LLM_TENSOR_FFN_UP,                                 "blk.%d.ffn_up" },
+    { LLM_TENSOR_FFN_GATE_EXP,                           "blk.%d.ffn_gate.%d" },
+    { LLM_TENSOR_FFN_DOWN_EXP,                           "blk.%d.ffn_down.%d" },
+    { LLM_TENSOR_FFN_UP_EXP,                             "blk.%d.ffn_up.%d" },
+    { LLM_TENSOR_FFN_GATE_EXPS,                          "blk.%d.ffn_gate_exps" },
+    { LLM_TENSOR_FFN_DOWN_EXPS,                          "blk.%d.ffn_down_exps" },
+    { LLM_TENSOR_FFN_UP_EXPS,                            "blk.%d.ffn_up_exps" },
+    { LLM_TENSOR_ATTN_POST_NORM,                         "blk.%d.post_attention_norm" },
+    { LLM_TENSOR_ATTN_Q_NORM,                            "blk.%d.attn_q_norm" },
+    { LLM_TENSOR_ATTN_K_NORM,                            "blk.%d.attn_k_norm" },
+    { LLM_TENSOR_ATTN_GATE,                              "blk.%d.attn_gate" },
+    { LLM_TENSOR_FFN_POST_NORM,                          "blk.%d.post_ffw_norm" },
+    { LLM_TENSOR_FFN_GATE_SHEXP,                         "blk.%d.ffn_gate_shexp" },
+    { LLM_TENSOR_FFN_UP_SHEXP,                           "blk.%d.ffn_up_shexp" },
+    { LLM_TENSOR_FFN_DOWN_SHEXP,                         "blk.%d.ffn_down_shexp" },
+    { LLM_TENSOR_FFN_EXP_PROBS_B,                        "blk.%d.exp_probs_b" },
+    { LLM_TENSOR_ATTN_NORM_2,                            "blk.%d.attn_norm_2" },
+    { LLM_TENSOR_ATTN_QKV,                               "blk.%d.attn_qkv" },
+    { LLM_TENSOR_LAYER_OUT_NORM,                         "blk.%d.layer_output_norm" },
+    { LLM_TENSOR_ATTN_OUT_NORM,                          "blk.%d.attn_output_norm" },
+    { LLM_TENSOR_POS_EMBD,                               "position_embd" },
+    { LLM_TENSOR_FFN_ACT,                                "blk.%d.ffn.act" },
+    { LLM_TENSOR_TOKEN_EMBD_NORM,                        "token_embd_norm" },
+    { LLM_TENSOR_TOKEN_TYPES,                            "token_types" },
+    { LLM_TENSOR_CLS,                                    "cls" },
+    { LLM_TENSOR_CLS_OUT,                                "cls.output" },
+    { LLM_TENSOR_ENC_OUTPUT_NORM,                        "enc.output_norm" },
+    { LLM_TENSOR_FFN_GATE_INP_SHEXP,                     "blk.%d.ffn_gate_inp_shexp" },
+    { LLM_TENSOR_SSM_A_NOSCAN,                           "blk.%d.ssm_a" },
+    { LLM_TENSOR_SSM_CONV1D,                             "blk.%d.ssm_conv1d" },
+    { LLM_TENSOR_SSM_DT,                                 "blk.%d.ssm_dt" },
+    { LLM_TENSOR_SSM_BETA_ALPHA,                         "blk.%d.ssm_ba" },
+    { LLM_TENSOR_SSM_IN,                                 "blk.%d.ssm_in" },
+    { LLM_TENSOR_SSM_NORM,                               "blk.%d.ssm_norm" },
+    { LLM_TENSOR_SSM_OUT,                                "blk.%d.ssm_out" },
+    { LLM_TENSOR_ROPE_FACTORS_LONG,                      "rope_factors_long" },
+    { LLM_TENSOR_ROPE_FACTORS_SHORT,                     "rope_factors_short" },
+    { LLM_TENSOR_SSM_X,                                  "blk.%d.ssm_x" },
+    { LLM_TENSOR_SSM_A,                                  "blk.%d.ssm_a" },
+    { LLM_TENSOR_SSM_D,                                  "blk.%d.ssm_d" },
+    { LLM_TENSOR_SSM_DT_NORM,                            "blk.%d.ssm_dt_norm" },
+    { LLM_TENSOR_SSM_B_NORM,                             "blk.%d.ssm_b_norm" },
+    { LLM_TENSOR_SSM_C_NORM,                             "blk.%d.ssm_c_norm" },
+    { LLM_TENSOR_ATTN_Q_A_NORM,                          "blk.%d.attn_q_a_norm" },
+    { LLM_TENSOR_ATTN_KV_A_NORM,                         "blk.%d.attn_kv_a_norm" },
+    { LLM_TENSOR_ATTN_Q_A,                               "blk.%d.attn_q_a" },
+    { LLM_TENSOR_ATTN_Q_B,                               "blk.%d.attn_q_b" },
+    { LLM_TENSOR_ATTN_KV_A_MQA,                          "blk.%d.attn_kv_a_mqa" },
+    { LLM_TENSOR_ATTN_KV_B,                              "blk.%d.attn_kv_b" },
+    { LLM_TENSOR_PER_LAYER_TOKEN_EMBD,                   "per_layer_token_embd" },
+    { LLM_TENSOR_PER_LAYER_MODEL_PROJ,                   "per_layer_model_proj" },
+    { LLM_TENSOR_PER_LAYER_PROJ_NORM,                    "per_layer_proj_norm" },
+    { LLM_TENSOR_ALTUP_UNEMBD_PROJ,                      "altup_unembd_proj" },
+    { LLM_TENSOR_ALTUP_PROJ,                             "altup_proj" },
+    { LLM_TENSOR_PER_LAYER_INP_GATE,                     "blk.%d.inp_gate" },
+    { LLM_TENSOR_PER_LAYER_PROJ,                         "blk.%d.proj" },
+    { LLM_TENSOR_PER_LAYER_POST_NORM,                    "blk.%d.post_norm" },
+    { LLM_TENSOR_ALTUP_CORRECT_COEF,                     "blk.%d.altup_correct_coef" },
+    { LLM_TENSOR_ALTUP_CORRECT_SCALE,                    "blk.%d.altup_correct_scale" },
+    { LLM_TENSOR_ALTUP_PREDICT_COEF,                     "blk.%d.altup_predict_coef" },
+    { LLM_TENSOR_ALTUP_ROUTER,                           "blk.%d.altup_router" },
+    { LLM_TENSOR_ALTUP_ROUTER_NORM,                      "blk.%d.altup_router_norm" },
+    { LLM_TENSOR_LAUREL_L,                               "blk.%d.laurel_l" },
+    { LLM_TENSOR_LAUREL_R,                               "blk.%d.laurel_r" },
+    { LLM_TENSOR_LAUREL_POST_NORM,                       "blk.%d.laurel_post_norm" },
+    { LLM_TENSOR_DENSE_2_OUT,                            "dense_2" },
+    { LLM_TENSOR_DENSE_3_OUT,                            "dense_3" },
+    { LLM_TENSOR_FFN_NORM_EXPS,                          "blk.%d.ffn_norm_exps" },
+    { LLM_TENSOR_ATTN_K_B,                               "blk.%d.attn_k_b" },
+    { LLM_TENSOR_ATTN_V_B,                               "blk.%d.attn_v_b" },
+    { LLM_TENSOR_NEXTN_EH_PROJ,                          "blk.%d.nextn.eh_proj" },
+    { LLM_TENSOR_NEXTN_EMBED_TOKENS,                     "blk.%d.nextn.embed_tokens" },
+    { LLM_TENSOR_NEXTN_ENORM,                            "blk.%d.nextn.enorm" },
+    { LLM_TENSOR_NEXTN_HNORM,                            "blk.%d.nextn.hnorm" },
+    { LLM_TENSOR_NEXTN_SHARED_HEAD_HEAD,                 "blk.%d.nextn.shared_head_head" },
+    { LLM_TENSOR_NEXTN_SHARED_HEAD_NORM,                 "blk.%d.nextn.shared_head_norm" },
+    { LLM_TENSOR_ATTN_SUB_NORM,                          "blk.%d.attn_sub_norm" },
+    { LLM_TENSOR_FFN_SUB_NORM,                           "blk.%d.ffn_sub_norm" },
+    { LLM_TENSOR_DEC_OUTPUT_NORM,                        "dec.output_norm" },
+    { LLM_TENSOR_DEC_ATTN_NORM,                          "dec.blk.%d.attn_norm" },
+    { LLM_TENSOR_DEC_ATTN_Q,                             "dec.blk.%d.attn_q" },
+    { LLM_TENSOR_DEC_ATTN_K,                             "dec.blk.%d.attn_k" },
+    { LLM_TENSOR_DEC_ATTN_V,                             "dec.blk.%d.attn_v" },
+    { LLM_TENSOR_DEC_ATTN_OUT,                           "dec.blk.%d.attn_o" },
+    { LLM_TENSOR_DEC_ATTN_REL_B,                         "dec.blk.%d.attn_rel_b" },
+    { LLM_TENSOR_DEC_CROSS_ATTN_NORM,                    "dec.blk.%d.cross_attn_norm" },
+    { LLM_TENSOR_DEC_CROSS_ATTN_Q,                       "dec.blk.%d.cross_attn_q" },
+    { LLM_TENSOR_DEC_CROSS_ATTN_K,                       "dec.blk.%d.cross_attn_k" },
+    { LLM_TENSOR_DEC_CROSS_ATTN_V,                       "dec.blk.%d.cross_attn_v" },
+    { LLM_TENSOR_DEC_CROSS_ATTN_OUT,                     "dec.blk.%d.cross_attn_o" },
+    { LLM_TENSOR_DEC_CROSS_ATTN_REL_B,                   "dec.blk.%d.cross_attn_rel_b" },
+    { LLM_TENSOR_DEC_FFN_NORM,                           "dec.blk.%d.ffn_norm" },
+    { LLM_TENSOR_DEC_FFN_GATE,                           "dec.blk.%d.ffn_gate" },
+    { LLM_TENSOR_DEC_FFN_DOWN,                           "dec.blk.%d.ffn_down" },
+    { LLM_TENSOR_DEC_FFN_UP,                             "dec.blk.%d.ffn_up" },
+    { LLM_TENSOR_ENC_ATTN_NORM,                          "enc.blk.%d.attn_norm" },
+    { LLM_TENSOR_ENC_ATTN_Q,                             "enc.blk.%d.attn_q" },
+    { LLM_TENSOR_ENC_ATTN_K,                             "enc.blk.%d.attn_k" },
+    { LLM_TENSOR_ENC_ATTN_V,                             "enc.blk.%d.attn_v" },
+    { LLM_TENSOR_ENC_ATTN_OUT,                           "enc.blk.%d.attn_o" },
+    { LLM_TENSOR_ENC_ATTN_REL_B,                         "enc.blk.%d.attn_rel_b" },
+    { LLM_TENSOR_ENC_FFN_NORM,                           "enc.blk.%d.ffn_norm" },
+    { LLM_TENSOR_ENC_FFN_GATE,                           "enc.blk.%d.ffn_gate" },
+    { LLM_TENSOR_ENC_FFN_DOWN,                           "enc.blk.%d.ffn_down" },
+    { LLM_TENSOR_ENC_FFN_UP,                             "enc.blk.%d.ffn_up" },
+    { LLM_TENSOR_TIME_MIX_W1,                            "blk.%d.time_mix_w1" },
+    { LLM_TENSOR_TIME_MIX_W2,                            "blk.%d.time_mix_w2" },
+    { LLM_TENSOR_TIME_MIX_LERP_X,                        "blk.%d.time_mix_lerp_x" },
+    { LLM_TENSOR_TIME_MIX_LERP_W,                        "blk.%d.time_mix_lerp_w" },
+    { LLM_TENSOR_TIME_MIX_LERP_K,                        "blk.%d.time_mix_lerp_k" },
+    { LLM_TENSOR_TIME_MIX_LERP_V,                        "blk.%d.time_mix_lerp_v" },
+    { LLM_TENSOR_TIME_MIX_LERP_R,                        "blk.%d.time_mix_lerp_r" },
+    { LLM_TENSOR_TIME_MIX_LERP_G,                        "blk.%d.time_mix_lerp_g" },
+    { LLM_TENSOR_TIME_MIX_LERP_FUSED,                    "blk.%d.time_mix_lerp_fused" },
+    { LLM_TENSOR_TIME_MIX_FIRST,                         "blk.%d.time_mix_first" },
+    { LLM_TENSOR_TIME_MIX_DECAY,                         "blk.%d.time_mix_decay" },
+    { LLM_TENSOR_TIME_MIX_DECAY_W1,                      "blk.%d.time_mix_decay_w1" },
+    { LLM_TENSOR_TIME_MIX_DECAY_W2,                      "blk.%d.time_mix_decay_w2" },
+    { LLM_TENSOR_TIME_MIX_KEY,                           "blk.%d.time_mix_key" },
+    { LLM_TENSOR_TIME_MIX_VALUE,                         "blk.%d.time_mix_value" },
+    { LLM_TENSOR_TIME_MIX_RECEPTANCE,                    "blk.%d.time_mix_receptance" },
+    { LLM_TENSOR_TIME_MIX_GATE,                          "blk.%d.time_mix_gate" },
+    { LLM_TENSOR_TIME_MIX_LN,                            "blk.%d.time_mix_ln" },
+    { LLM_TENSOR_TIME_MIX_OUTPUT,                        "blk.%d.time_mix_output" },
+    { LLM_TENSOR_CHANNEL_MIX_LERP_K,                     "blk.%d.channel_mix_lerp_k" },
+    { LLM_TENSOR_CHANNEL_MIX_LERP_R,                     "blk.%d.channel_mix_lerp_r" },
+    { LLM_TENSOR_CHANNEL_MIX_KEY,                        "blk.%d.channel_mix_key" },
+    { LLM_TENSOR_CHANNEL_MIX_VALUE,                      "blk.%d.channel_mix_value" },
+    { LLM_TENSOR_CHANNEL_MIX_RECEPTANCE,                 "blk.%d.channel_mix_receptance" },
+    { LLM_TENSOR_TIME_MIX_W0,                            "blk.%d.time_mix_w0" },
+    { LLM_TENSOR_TIME_MIX_A0,                            "blk.%d.time_mix_a0" },
+    { LLM_TENSOR_TIME_MIX_A1,                            "blk.%d.time_mix_a1" },
+    { LLM_TENSOR_TIME_MIX_A2,                            "blk.%d.time_mix_a2" },
+    { LLM_TENSOR_TIME_MIX_V0,                            "blk.%d.time_mix_v0" },
+    { LLM_TENSOR_TIME_MIX_V1,                            "blk.%d.time_mix_v1" },
+    { LLM_TENSOR_TIME_MIX_V2,                            "blk.%d.time_mix_v2" },
+    { LLM_TENSOR_TIME_MIX_G1,                            "blk.%d.time_mix_g1" },
+    { LLM_TENSOR_TIME_MIX_G2,                            "blk.%d.time_mix_g2" },
+    { LLM_TENSOR_TIME_MIX_K_K,                           "blk.%d.time_mix_k_k" },
+    { LLM_TENSOR_TIME_MIX_K_A,                           "blk.%d.time_mix_k_a" },
+    { LLM_TENSOR_TIME_MIX_R_K,                           "blk.%d.time_mix_r_k" },
+    { LLM_TENSOR_CONV1D,                                 "conv1d" },
+    { LLM_TENSOR_CONVNEXT_DW,                            "convnext.%d.dw" },
+    { LLM_TENSOR_CONVNEXT_NORM,                          "convnext.%d.norm" },
+    { LLM_TENSOR_CONVNEXT_PW1,                           "convnext.%d.pw1" },
+    { LLM_TENSOR_CONVNEXT_PW2,                           "convnext.%d.pw2" },
+    { LLM_TENSOR_CONVNEXT_GAMMA,                         "convnext.%d.gamma" },
+    { LLM_TENSOR_POS_NET_CONV1,                          "posnet.%d.conv1" },
+    { LLM_TENSOR_POS_NET_CONV2,                          "posnet.%d.conv2" },
+    { LLM_TENSOR_POS_NET_NORM,                           "posnet.%d.norm" },
+    { LLM_TENSOR_POS_NET_NORM1,                          "posnet.%d.norm1" },
+    { LLM_TENSOR_POS_NET_NORM2,                          "posnet.%d.norm2" },
+    { LLM_TENSOR_POS_NET_ATTN_NORM,                      "posnet.%d.attn_norm" },
+    { LLM_TENSOR_POS_NET_ATTN_Q,                         "posnet.%d.attn_q" },
+    { LLM_TENSOR_POS_NET_ATTN_K,                         "posnet.%d.attn_k" },
+    { LLM_TENSOR_POS_NET_ATTN_V,                         "posnet.%d.attn_v" },
+    { LLM_TENSOR_POS_NET_ATTN_OUT,                       "posnet.%d.attn_output" },
+    { LLM_TENSOR_ATTN_SINKS,                             "blk.%d.attn_sinks" },
+    { LLM_TENSOR_SHORTCONV_CONV,                         "blk.%d.shortconv.conv" },
+    { LLM_TENSOR_SHORTCONV_INPROJ,                       "blk.%d.shortconv.in_proj" },
+    { LLM_TENSOR_SHORTCONV_OUTPROJ,                      "blk.%d.shortconv.out_proj" },
+    { LLM_TENSOR_FFN_GATE_CHEXPS,                        "blk.%d.ffn_gate_chexps" },
+    { LLM_TENSOR_FFN_DOWN_CHEXPS,                        "blk.%d.ffn_down_chexps" },
+    { LLM_TENSOR_FFN_UP_CHEXPS,                          "blk.%d.ffn_up_chexps" },
+    { LLM_TENSOR_VISEXP_ATTN_QKV,                        "blk.%d.vis_attn_qkv" },
+    { LLM_TENSOR_VISEXP_ATTN_OUT,                        "blk.%d.vis_attn_output" },
+    { LLM_TENSOR_VISEXP_FFN_GATE,                        "blk.%d.vis_gate" },
+    { LLM_TENSOR_VISEXP_FFN_DOWN,                        "blk.%d.vis_down" },
+    { LLM_TENSOR_VISEXP_FFN_UP,                          "blk.%d.vis_up" },
 };
 
+static std::set<llm_tensor> llm_get_tensor_names(llm_arch arch) {
+    switch (arch) {
+        case LLM_ARCH_CLIP:
+            return {};
+        case LLM_ARCH_LLAMA:
+        case LLM_ARCH_DECI:
+        case LLM_ARCH_MISTRAL3:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_ROT_EMBD,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE_EXP,
+                LLM_TENSOR_FFN_DOWN_EXP,
+                LLM_TENSOR_FFN_UP_EXP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+            };
+        case LLM_ARCH_ARCEE:
+        case LLM_ARCH_STARCODER2:
+        case LLM_ARCH_NEMOTRON:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_ROT_EMBD,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_AFMOE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_GATE,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_POST_NORM,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_EXP_PROBS_B,
+            };
+        case LLM_ARCH_LLAMA4:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_ROT_EMBD,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE_EXP,
+                LLM_TENSOR_FFN_DOWN_EXP,
+                LLM_TENSOR_FFN_UP_EXP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+            };
+        case LLM_ARCH_BAICHUAN:
+        case LLM_ARCH_ORION:
+        case LLM_ARCH_XVERSE:
+        case LLM_ARCH_EXAONE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_ROT_EMBD,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_FALCON:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_NORM_2,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_GROK:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_ROT_EMBD,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE_EXP,
+                LLM_TENSOR_FFN_DOWN_EXP,
+                LLM_TENSOR_FFN_UP_EXP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_POST_NORM,
+                LLM_TENSOR_LAYER_OUT_NORM,
+                LLM_TENSOR_ATTN_OUT_NORM,
+            };
+        case LLM_ARCH_GPT2:
+        case LLM_ARCH_STARCODER:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_POS_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_DOWN,
+            };
+        case LLM_ARCH_GPTNEOX:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_MPT:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_ACT,
+                LLM_TENSOR_POS_EMBD,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K_NORM,
+            };
+        case LLM_ARCH_REFACT:
+        case LLM_ARCH_QWEN2:
+        case LLM_ARCH_QWEN2VL:
+        case LLM_ARCH_INTERNLM2:
+        case LLM_ARCH_GRANITE:
+        case LLM_ARCH_ERNIE4_5:
+        case LLM_ARCH_SMOLLM3:
+        case LLM_ARCH_DREAM:
+        case LLM_ARCH_LLADA:
+        case LLM_ARCH_PANGU_EMBED:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_BERT:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_TOKEN_EMBD_NORM,
+                LLM_TENSOR_TOKEN_TYPES,
+                LLM_TENSOR_POS_EMBD,
+                LLM_TENSOR_ATTN_OUT_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_LAYER_OUT_NORM,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_CLS,
+                LLM_TENSOR_CLS_OUT,
+            };
+        case LLM_ARCH_NOMIC_BERT:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_TOKEN_EMBD_NORM,
+                LLM_TENSOR_TOKEN_TYPES,
+                LLM_TENSOR_ATTN_OUT_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_LAYER_OUT_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_NOMIC_BERT_MOE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_TOKEN_EMBD_NORM,
+                LLM_TENSOR_TOKEN_TYPES,
+                LLM_TENSOR_ATTN_OUT_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_LAYER_OUT_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+            };
+        case LLM_ARCH_NEO_BERT:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_ENC_OUTPUT_NORM,
+                LLM_TENSOR_CLS,
+                LLM_TENSOR_CLS_OUT,
+            };
+        case LLM_ARCH_JINA_BERT_V2:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_TOKEN_EMBD_NORM,
+                LLM_TENSOR_TOKEN_TYPES,
+                LLM_TENSOR_ATTN_NORM_2,
+                LLM_TENSOR_ATTN_OUT_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_LAYER_OUT_NORM,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_CLS,
+            };
+        case LLM_ARCH_JINA_BERT_V3:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_TOKEN_EMBD_NORM,
+                LLM_TENSOR_TOKEN_TYPES,
+                LLM_TENSOR_ATTN_OUT_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_LAYER_OUT_NORM,
+            };
+        case LLM_ARCH_BLOOM:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_TOKEN_EMBD_NORM,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_DOWN,
+            };
+        case LLM_ARCH_STABLELM:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K_NORM,
+            };
+        case LLM_ARCH_QWEN:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_QWEN2MOE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_INP_SHEXP,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+            };
+        case LLM_ARCH_QWEN3:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_CLS_OUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_QWEN3MOE:
+        case LLM_ARCH_QWEN3VLMOE:
+        case LLM_ARCH_OLMOE:
+        case LLM_ARCH_LLADA_MOE:
+        case LLM_ARCH_RND1:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+            };
+        case LLM_ARCH_QWEN3NEXT:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_INP_SHEXP,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+                LLM_TENSOR_SSM_A_NOSCAN,
+                LLM_TENSOR_SSM_CONV1D,
+                LLM_TENSOR_SSM_DT,
+                LLM_TENSOR_SSM_BETA_ALPHA,
+                LLM_TENSOR_SSM_IN,
+                LLM_TENSOR_SSM_NORM,
+                LLM_TENSOR_SSM_OUT,
+            };
+        case LLM_ARCH_QWEN3VL:
+        case LLM_ARCH_CHAMELEON:
+        case LLM_ARCH_HUNYUAN_DENSE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_PHI2:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_PHI3:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FACTORS_LONG,
+                LLM_TENSOR_ROPE_FACTORS_SHORT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_PHIMOE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FACTORS_LONG,
+                LLM_TENSOR_ROPE_FACTORS_SHORT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+            };
+        case LLM_ARCH_PLAMO:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_ROT_EMBD,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_PLAMO2:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_ROT_EMBD,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_SSM_IN,
+                LLM_TENSOR_SSM_CONV1D,
+                LLM_TENSOR_SSM_X,
+                LLM_TENSOR_SSM_DT,
+                LLM_TENSOR_SSM_A,
+                LLM_TENSOR_SSM_D,
+                LLM_TENSOR_SSM_OUT,
+                LLM_TENSOR_SSM_DT_NORM,
+                LLM_TENSOR_SSM_B_NORM,
+                LLM_TENSOR_SSM_C_NORM,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_FFN_POST_NORM,
+            };
+        case LLM_ARCH_CODESHELL:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_ROT_EMBD,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_MINICPM:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ROPE_FACTORS_LONG,
+                LLM_TENSOR_ROPE_FACTORS_SHORT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_ROT_EMBD,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE_EXP,
+                LLM_TENSOR_FFN_DOWN_EXP,
+                LLM_TENSOR_FFN_UP_EXP,
+            };
+        case LLM_ARCH_MINICPM3:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FACTORS_LONG,
+                LLM_TENSOR_ROPE_FACTORS_SHORT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q_A_NORM,
+                LLM_TENSOR_ATTN_KV_A_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_A,
+                LLM_TENSOR_ATTN_Q_B,
+                LLM_TENSOR_ATTN_KV_A_MQA,
+                LLM_TENSOR_ATTN_KV_B,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_DOWN,
+            };
+        case LLM_ARCH_GEMMA:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_GEMMA2:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_POST_NORM,
+            };
+        case LLM_ARCH_GEMMA3:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_POST_NORM,
+            };
+        case LLM_ARCH_GEMMA3N:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_POST_NORM,
+                LLM_TENSOR_PER_LAYER_TOKEN_EMBD,
+                LLM_TENSOR_PER_LAYER_MODEL_PROJ,
+                LLM_TENSOR_PER_LAYER_PROJ_NORM,
+                LLM_TENSOR_ALTUP_UNEMBD_PROJ,
+                LLM_TENSOR_ALTUP_PROJ,
+                LLM_TENSOR_PER_LAYER_INP_GATE,
+                LLM_TENSOR_PER_LAYER_PROJ,
+                LLM_TENSOR_PER_LAYER_POST_NORM,
+                LLM_TENSOR_ALTUP_CORRECT_COEF,
+                LLM_TENSOR_ALTUP_CORRECT_SCALE,
+                LLM_TENSOR_ALTUP_PREDICT_COEF,
+                LLM_TENSOR_ALTUP_ROUTER,
+                LLM_TENSOR_ALTUP_ROUTER_NORM,
+                LLM_TENSOR_LAUREL_L,
+                LLM_TENSOR_LAUREL_R,
+                LLM_TENSOR_LAUREL_POST_NORM,
+            };
+        case LLM_ARCH_GEMMA_EMBEDDING:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_DENSE_2_OUT,
+                LLM_TENSOR_DENSE_3_OUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_POST_NORM,
+            };
+        case LLM_ARCH_MAMBA:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_SSM_IN,
+                LLM_TENSOR_SSM_CONV1D,
+                LLM_TENSOR_SSM_X,
+                LLM_TENSOR_SSM_DT,
+                LLM_TENSOR_SSM_A,
+                LLM_TENSOR_SSM_D,
+                LLM_TENSOR_SSM_OUT,
+            };
+        case LLM_ARCH_MAMBA2:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_SSM_IN,
+                LLM_TENSOR_SSM_CONV1D,
+                LLM_TENSOR_SSM_DT,
+                LLM_TENSOR_SSM_A,
+                LLM_TENSOR_SSM_D,
+                LLM_TENSOR_SSM_NORM,
+                LLM_TENSOR_SSM_OUT,
+            };
+        case LLM_ARCH_JAMBA:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_SSM_IN,
+                LLM_TENSOR_SSM_CONV1D,
+                LLM_TENSOR_SSM_X,
+                LLM_TENSOR_SSM_DT,
+                LLM_TENSOR_SSM_DT_NORM,
+                LLM_TENSOR_SSM_A,
+                LLM_TENSOR_SSM_B_NORM,
+                LLM_TENSOR_SSM_C_NORM,
+                LLM_TENSOR_SSM_D,
+                LLM_TENSOR_SSM_OUT,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+            };
+        case LLM_ARCH_FALCON_H1:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_SSM_IN,
+                LLM_TENSOR_SSM_CONV1D,
+                LLM_TENSOR_SSM_DT,
+                LLM_TENSOR_SSM_A,
+                LLM_TENSOR_SSM_D,
+                LLM_TENSOR_SSM_NORM,
+                LLM_TENSOR_SSM_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_COMMAND_R:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K_NORM,
+            };
+        case LLM_ARCH_COHERE2:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_DBRX:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_OUT_NORM,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+            };
+        case LLM_ARCH_OLMO:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_OLMO2:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_FFN_POST_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_OPENELM:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_ARCTIC:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_NORM_EXPS,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+            };
+        case LLM_ARCH_DEEPSEEK:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_ROT_EMBD,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_INP_SHEXP,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+            };
+        case LLM_ARCH_DEEPSEEK2:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q_A_NORM,
+                LLM_TENSOR_ATTN_KV_A_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_A,
+                LLM_TENSOR_ATTN_Q_B,
+                LLM_TENSOR_ATTN_KV_A_MQA,
+                LLM_TENSOR_ATTN_KV_B,
+                LLM_TENSOR_ATTN_K_B,
+                LLM_TENSOR_ATTN_V_B,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_INP_SHEXP,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+                LLM_TENSOR_FFN_EXP_PROBS_B,
+            };
+        case LLM_ARCH_PLM:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_KV_A_MQA,
+                LLM_TENSOR_ATTN_KV_A_NORM,
+                LLM_TENSOR_ATTN_KV_B,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_CHATGLM:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_DOWN,
+            };
+        case LLM_ARCH_GLM4:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_FFN_POST_NORM,
+            };
+        case LLM_ARCH_GLM4_MOE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+                LLM_TENSOR_FFN_EXP_PROBS_B,
+                LLM_TENSOR_NEXTN_EH_PROJ,
+                LLM_TENSOR_NEXTN_EMBED_TOKENS,
+                LLM_TENSOR_NEXTN_ENORM,
+                LLM_TENSOR_NEXTN_HNORM,
+                LLM_TENSOR_NEXTN_SHARED_HEAD_HEAD,
+                LLM_TENSOR_NEXTN_SHARED_HEAD_NORM,
+            };
+        case LLM_ARCH_BITNET:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_SUB_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_SUB_NORM,
+            };
+        case LLM_ARCH_T5:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_DEC_OUTPUT_NORM,
+                LLM_TENSOR_DEC_ATTN_NORM,
+                LLM_TENSOR_DEC_ATTN_Q,
+                LLM_TENSOR_DEC_ATTN_K,
+                LLM_TENSOR_DEC_ATTN_V,
+                LLM_TENSOR_DEC_ATTN_OUT,
+                LLM_TENSOR_DEC_ATTN_REL_B,
+                LLM_TENSOR_DEC_CROSS_ATTN_NORM,
+                LLM_TENSOR_DEC_CROSS_ATTN_Q,
+                LLM_TENSOR_DEC_CROSS_ATTN_K,
+                LLM_TENSOR_DEC_CROSS_ATTN_V,
+                LLM_TENSOR_DEC_CROSS_ATTN_OUT,
+                LLM_TENSOR_DEC_CROSS_ATTN_REL_B,
+                LLM_TENSOR_DEC_FFN_NORM,
+                LLM_TENSOR_DEC_FFN_GATE,
+                LLM_TENSOR_DEC_FFN_DOWN,
+                LLM_TENSOR_DEC_FFN_UP,
+                LLM_TENSOR_ENC_OUTPUT_NORM,
+                LLM_TENSOR_ENC_ATTN_NORM,
+                LLM_TENSOR_ENC_ATTN_Q,
+                LLM_TENSOR_ENC_ATTN_K,
+                LLM_TENSOR_ENC_ATTN_V,
+                LLM_TENSOR_ENC_ATTN_OUT,
+                LLM_TENSOR_ENC_ATTN_REL_B,
+                LLM_TENSOR_ENC_FFN_NORM,
+                LLM_TENSOR_ENC_FFN_GATE,
+                LLM_TENSOR_ENC_FFN_DOWN,
+                LLM_TENSOR_ENC_FFN_UP,
+            };
+        case LLM_ARCH_T5ENCODER:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ENC_OUTPUT_NORM,
+                LLM_TENSOR_ENC_ATTN_NORM,
+                LLM_TENSOR_ENC_ATTN_Q,
+                LLM_TENSOR_ENC_ATTN_K,
+                LLM_TENSOR_ENC_ATTN_V,
+                LLM_TENSOR_ENC_ATTN_OUT,
+                LLM_TENSOR_ENC_ATTN_REL_B,
+                LLM_TENSOR_ENC_FFN_NORM,
+                LLM_TENSOR_ENC_FFN_GATE,
+                LLM_TENSOR_ENC_FFN_DOWN,
+                LLM_TENSOR_ENC_FFN_UP,
+            };
+        case LLM_ARCH_JAIS:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+            };
+        case LLM_ARCH_NEMOTRON_H:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_SSM_IN,
+                LLM_TENSOR_SSM_CONV1D,
+                LLM_TENSOR_SSM_DT,
+                LLM_TENSOR_SSM_A,
+                LLM_TENSOR_SSM_D,
+                LLM_TENSOR_SSM_NORM,
+                LLM_TENSOR_SSM_OUT,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_NEMOTRON_H_MOE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                // mamba(2) ssm layers
+                LLM_TENSOR_SSM_IN,
+                LLM_TENSOR_SSM_CONV1D,
+                LLM_TENSOR_SSM_DT,
+                LLM_TENSOR_SSM_A,
+                LLM_TENSOR_SSM_D,
+                LLM_TENSOR_SSM_NORM,
+                LLM_TENSOR_SSM_OUT,
+                // attention layers
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                // dense FFN
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                // MoE FFN (for MoE layers)
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_EXP_PROBS_B,
+                // MoE shared expert layer
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+            };
+        case LLM_ARCH_EXAONE4:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_POST_NORM,
+            };
+        case LLM_ARCH_RWKV6:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_TOKEN_EMBD_NORM,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_NORM_2,
+                LLM_TENSOR_TIME_MIX_W1,
+                LLM_TENSOR_TIME_MIX_W2,
+                LLM_TENSOR_TIME_MIX_LERP_X,
+                LLM_TENSOR_TIME_MIX_LERP_W,
+                LLM_TENSOR_TIME_MIX_LERP_K,
+                LLM_TENSOR_TIME_MIX_LERP_V,
+                LLM_TENSOR_TIME_MIX_LERP_R,
+                LLM_TENSOR_TIME_MIX_LERP_G,
+                LLM_TENSOR_TIME_MIX_LERP_FUSED,
+                LLM_TENSOR_TIME_MIX_FIRST,
+                LLM_TENSOR_TIME_MIX_DECAY,
+                LLM_TENSOR_TIME_MIX_DECAY_W1,
+                LLM_TENSOR_TIME_MIX_DECAY_W2,
+                LLM_TENSOR_TIME_MIX_KEY,
+                LLM_TENSOR_TIME_MIX_VALUE,
+                LLM_TENSOR_TIME_MIX_RECEPTANCE,
+                LLM_TENSOR_TIME_MIX_GATE,
+                LLM_TENSOR_TIME_MIX_LN,
+                LLM_TENSOR_TIME_MIX_OUTPUT,
+                LLM_TENSOR_CHANNEL_MIX_LERP_K,
+                LLM_TENSOR_CHANNEL_MIX_LERP_R,
+                LLM_TENSOR_CHANNEL_MIX_KEY,
+                LLM_TENSOR_CHANNEL_MIX_VALUE,
+                LLM_TENSOR_CHANNEL_MIX_RECEPTANCE,
+            };
+        case LLM_ARCH_RWKV6QWEN2:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_TIME_MIX_W1,
+                LLM_TENSOR_TIME_MIX_W2,
+                LLM_TENSOR_TIME_MIX_LERP_X,
+                LLM_TENSOR_TIME_MIX_LERP_FUSED,
+                LLM_TENSOR_TIME_MIX_FIRST,
+                LLM_TENSOR_TIME_MIX_DECAY,
+                LLM_TENSOR_TIME_MIX_DECAY_W1,
+                LLM_TENSOR_TIME_MIX_DECAY_W2,
+                LLM_TENSOR_TIME_MIX_KEY,
+                LLM_TENSOR_TIME_MIX_VALUE,
+                LLM_TENSOR_TIME_MIX_RECEPTANCE,
+                LLM_TENSOR_TIME_MIX_GATE,
+                LLM_TENSOR_TIME_MIX_OUTPUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_RWKV7:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_TOKEN_EMBD_NORM,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_NORM_2,
+                LLM_TENSOR_TIME_MIX_W0,
+                LLM_TENSOR_TIME_MIX_W1,
+                LLM_TENSOR_TIME_MIX_W2,
+                LLM_TENSOR_TIME_MIX_A0,
+                LLM_TENSOR_TIME_MIX_A1,
+                LLM_TENSOR_TIME_MIX_A2,
+                LLM_TENSOR_TIME_MIX_V0,
+                LLM_TENSOR_TIME_MIX_V1,
+                LLM_TENSOR_TIME_MIX_V2,
+                LLM_TENSOR_TIME_MIX_G1,
+                LLM_TENSOR_TIME_MIX_G2,
+                LLM_TENSOR_TIME_MIX_K_K,
+                LLM_TENSOR_TIME_MIX_K_A,
+                LLM_TENSOR_TIME_MIX_R_K,
+                LLM_TENSOR_TIME_MIX_LERP_FUSED,
+                LLM_TENSOR_TIME_MIX_KEY,
+                LLM_TENSOR_TIME_MIX_VALUE,
+                LLM_TENSOR_TIME_MIX_RECEPTANCE,
+                LLM_TENSOR_TIME_MIX_LN,
+                LLM_TENSOR_TIME_MIX_OUTPUT,
+                LLM_TENSOR_CHANNEL_MIX_LERP_K,
+                LLM_TENSOR_CHANNEL_MIX_KEY,
+                LLM_TENSOR_CHANNEL_MIX_VALUE,
+            };
+        case LLM_ARCH_ARWKV7:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_TOKEN_EMBD_NORM,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_TIME_MIX_W0,
+                LLM_TENSOR_TIME_MIX_W1,
+                LLM_TENSOR_TIME_MIX_W2,
+                LLM_TENSOR_TIME_MIX_A0,
+                LLM_TENSOR_TIME_MIX_A1,
+                LLM_TENSOR_TIME_MIX_A2,
+                LLM_TENSOR_TIME_MIX_V0,
+                LLM_TENSOR_TIME_MIX_V1,
+                LLM_TENSOR_TIME_MIX_V2,
+                LLM_TENSOR_TIME_MIX_G1,
+                LLM_TENSOR_TIME_MIX_G2,
+                LLM_TENSOR_TIME_MIX_K_K,
+                LLM_TENSOR_TIME_MIX_K_A,
+                LLM_TENSOR_TIME_MIX_R_K,
+                LLM_TENSOR_TIME_MIX_LERP_FUSED,
+                LLM_TENSOR_TIME_MIX_KEY,
+                LLM_TENSOR_TIME_MIX_VALUE,
+                LLM_TENSOR_TIME_MIX_RECEPTANCE,
+                LLM_TENSOR_TIME_MIX_LN,
+                LLM_TENSOR_TIME_MIX_OUTPUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_GRANITE_MOE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+            };
+        case LLM_ARCH_GRANITE_HYBRID:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_SSM_IN,
+                LLM_TENSOR_SSM_CONV1D,
+                LLM_TENSOR_SSM_DT,
+                LLM_TENSOR_SSM_A,
+                LLM_TENSOR_SSM_D,
+                LLM_TENSOR_SSM_NORM,
+                LLM_TENSOR_SSM_OUT,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+            };
+        case LLM_ARCH_WAVTOKENIZER_DEC:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_TOKEN_EMBD_NORM,
+                LLM_TENSOR_CONV1D,
+                LLM_TENSOR_CONVNEXT_DW,
+                LLM_TENSOR_CONVNEXT_NORM,
+                LLM_TENSOR_CONVNEXT_PW1,
+                LLM_TENSOR_CONVNEXT_PW2,
+                LLM_TENSOR_CONVNEXT_GAMMA,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_POS_NET_CONV1,
+                LLM_TENSOR_POS_NET_CONV2,
+                LLM_TENSOR_POS_NET_NORM,
+                LLM_TENSOR_POS_NET_NORM1,
+                LLM_TENSOR_POS_NET_NORM2,
+                LLM_TENSOR_POS_NET_ATTN_NORM,
+                LLM_TENSOR_POS_NET_ATTN_Q,
+                LLM_TENSOR_POS_NET_ATTN_K,
+                LLM_TENSOR_POS_NET_ATTN_V,
+                LLM_TENSOR_POS_NET_ATTN_OUT,
+            };
+        case LLM_ARCH_BAILINGMOE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_INP_SHEXP,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+            };
+        case LLM_ARCH_BAILINGMOE2:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_EXP_PROBS_B,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+                LLM_TENSOR_NEXTN_EH_PROJ,
+                LLM_TENSOR_NEXTN_EMBED_TOKENS,
+                LLM_TENSOR_NEXTN_ENORM,
+                LLM_TENSOR_NEXTN_HNORM,
+                LLM_TENSOR_NEXTN_SHARED_HEAD_HEAD,
+                LLM_TENSOR_NEXTN_SHARED_HEAD_NORM,
+                LLM_TENSOR_LAYER_OUT_NORM,
+            };
+        case LLM_ARCH_DOTS1:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_INP_SHEXP,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+                LLM_TENSOR_FFN_EXP_PROBS_B,
+            };
+        case LLM_ARCH_ERNIE4_5_MOE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_EXP_PROBS_B,
+            };
+        case LLM_ARCH_HUNYUAN_MOE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+            };
+        case LLM_ARCH_OPENAI_MOE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_SINKS,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+            };
+        case LLM_ARCH_LFM2:
+            return {
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_SHORTCONV_CONV,
+                LLM_TENSOR_SHORTCONV_INPROJ,
+                LLM_TENSOR_SHORTCONV_OUTPROJ,
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM_LFM2,
+                LLM_TENSOR_OUTPUT,
+            };
+        case LLM_ARCH_LFM2MOE:
+            return {
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_SHORTCONV_CONV,
+                LLM_TENSOR_SHORTCONV_INPROJ,
+                LLM_TENSOR_SHORTCONV_OUTPROJ,
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_EXP_PROBS_B,
+            };
+        case LLM_ARCH_SMALLTHINKER:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+            };
+        case LLM_ARCH_APERTUS:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_SEED_OSS:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_GROVEMOE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_CHEXPS,
+                LLM_TENSOR_FFN_DOWN_CHEXPS,
+                LLM_TENSOR_FFN_UP_CHEXPS,
+            };
+        case LLM_ARCH_MINIMAX_M2:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_EXP_PROBS_B,
+            };
+        case LLM_ARCH_COGVLM:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_VISEXP_ATTN_QKV,
+                LLM_TENSOR_VISEXP_ATTN_OUT,
+                LLM_TENSOR_VISEXP_FFN_GATE,
+                LLM_TENSOR_VISEXP_FFN_DOWN,
+                LLM_TENSOR_VISEXP_FFN_UP,
+            };
+        case LLM_ARCH_GPTJ:
+        case LLM_ARCH_UNKNOWN:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+            };
+        default:
+            GGML_ABORT("unknown architecture for tensor mapping");
+    }
+}
+
 // declare information about the model weight tensors:
 // - the layer in which the tensor is going to be used. this is needed in order to assign the correct buffer type for the weight
 // - the operator which is going to use the weight. this is needed to determine if the respective backend supports the operator
@@ -2603,6 +2202,7 @@ static const std::map<llm_tensor, llm_tensor_info> LLM_TENSOR_INFOS = {
     {LLM_TENSOR_DENSE_2_OUT,                {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL_MAT}}, // Dense layer output
     {LLM_TENSOR_DENSE_3_OUT,                {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL_MAT}}, // Dense layer output
     {LLM_TENSOR_OUTPUT_NORM,                {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL}},
+    {LLM_TENSOR_OUTPUT_NORM_LFM2,           {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL}},
     {LLM_TENSOR_DEC_OUTPUT_NORM,            {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL}},
     {LLM_TENSOR_ENC_OUTPUT_NORM,            {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL}},
     {LLM_TENSOR_ROPE_FREQS,                 {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ROPE}},
@@ -2791,13 +2391,20 @@ std::string LLM_KV::operator()(llm_kv kv) const {
     return name;
 }
 
+LLM_TN_IMPL::LLM_TN_IMPL(llm_arch arch, llm_tensor tensor, const char * suffix, int bid, int xid)
+    : arch(arch), tensor(tensor), suffix(suffix), bid(bid), xid(xid),
+      model_tensors(llm_get_tensor_names(arch)) {}
+
 std::string LLM_TN_IMPL::str() const {
-    if (LLM_TENSOR_NAMES.at(arch).find(tensor) == LLM_TENSOR_NAMES.at(arch).end()) {
-        return "__missing__";
+    if (LLM_TENSOR_NAMES.find(tensor) == LLM_TENSOR_NAMES.end()) {
+        GGML_ABORT("unknown tensor name for tensor id %d", static_cast<int>(tensor));
     }
 
-    std::string name = ::format(LLM_TENSOR_NAMES.at(arch).at(tensor), bid, xid);
+    if (model_tensors.find(tensor) == model_tensors.end()) {
+        return LLM_TENSOR_NAMES.at(tensor);
+    }
 
+    std::string name = ::format(LLM_TENSOR_NAMES.at(tensor), bid, xid);
     if (suffix != nullptr) {
         name += ".";
         name += suffix;
diff --git a/src/llama-arch.h b/src/llama-arch.h
index 455658f5dc..6cbf9b1f89 100644
--- a/src/llama-arch.h
+++ b/src/llama-arch.h
@@ -3,6 +3,7 @@
 #include "ggml.h" // ggml_op
 
 #include <string>
+#include <set>
 
 //
 // gguf constants (sync with gguf.py)
@@ -316,6 +317,7 @@ enum llm_tensor {
     LLM_TENSOR_DENSE_3_OUT,
     LLM_TENSOR_OUTPUT,
     LLM_TENSOR_OUTPUT_NORM,
+    LLM_TENSOR_OUTPUT_NORM_LFM2, // fix for wrong tensor name
     LLM_TENSOR_ROPE_FREQS,
     LLM_TENSOR_ROPE_FACTORS_LONG,
     LLM_TENSOR_ROPE_FACTORS_SHORT,
@@ -526,6 +528,10 @@ struct LLM_TN_IMPL {
     const int bid;
     const int xid;
 
+    const std::set<llm_tensor> model_tensors;
+
+    LLM_TN_IMPL(llm_arch arch, llm_tensor tensor, const char * suffix, int bid, int xid);
+
     std::string str() const;
 
     operator std::string() const {
@@ -547,11 +553,11 @@ struct LLM_TN {
     llm_arch arch;
 
     LLM_TN_IMPL operator()(llm_tensor tensor, const char * suffix, int bid = -1, int xid = -1) const {
-        return { arch, tensor, suffix, bid, xid };
+        return LLM_TN_IMPL(arch, tensor, suffix, bid, xid);
     }
 
     LLM_TN_IMPL operator()(llm_tensor tensor, int bid = -1, int xid = -1) const {
-        return { arch, tensor, nullptr, bid, xid };
+        return LLM_TN_IMPL(arch, tensor, nullptr, bid, xid);
     }
 };
 

From 79dbae034afdfaa8e17989ea1b9b20094c8d0a36 Mon Sep 17 00:00:00 2001
From: Daniel Bevenius <daniel.bevenius@gmail.com>
Date: Tue, 16 Dec 2025 13:25:09 +0100
Subject: [PATCH 4/6] model-conversion : remove -fa option in model card
 template [no ci] (#18088)

This commit updates the causal model card template and removes the
-fa option as it is no longer required (fa is auto detected).
---
 examples/model-conversion/scripts/causal/modelcard.template | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/examples/model-conversion/scripts/causal/modelcard.template b/examples/model-conversion/scripts/causal/modelcard.template
index 87800a1b93..cfa8e6b433 100644
--- a/examples/model-conversion/scripts/causal/modelcard.template
+++ b/examples/model-conversion/scripts/causal/modelcard.template
@@ -7,7 +7,7 @@ base_model:
 Recommended way to run this model:
 
 ```sh
-llama-server -hf {namespace}/{model_name}-GGUF -c 0 -fa
+llama-server -hf {namespace}/{model_name}-GGUF -c 0
 ```
 
 Then, access http://localhost:8080

From 59977eba7b0a3603d0017717d3beec7bde018f3c Mon Sep 17 00:00:00 2001
From: yifant-code <tian.yifan123@gmail.com>
Date: Tue, 16 Dec 2025 07:27:36 -0500
Subject: [PATCH 5/6] server: fix crash when batch > ubatch with embeddings
 (#17912)

* server: fix crash when batch > ubatch with embeddings (#12836)

Fixes #12836 where the server crashes with GGML_ASSERT failure when
running with embeddings enabled and n_batch > n_ubatch.

Root cause: Embeddings use non-causal attention which requires all
tokens to be processed within a single ubatch. When n_batch > n_ubatch,
the server attempts to split processing, causing assertion failure.

Solution:
- Add parameter validation in main() after common_params_parse()
- When embeddings enabled and n_batch > n_ubatch:
  * Log warnings explaining the issue
  * Automatically set n_batch = n_ubatch
  * Prevent server crash

This follows the approach suggested by @ggerganov in issue #12836.

Note: This supersedes stalled PR #12940 which attempted a runtime fix
in the old examples/server/server.cpp location. This implementation
validates at startup in tools/server/server.cpp (current location).

Testing:
- Build: Compiles successfully
- Validation triggers: Warns when -b > -ub with --embedding
- Auto-correction works: Adjusts n_batch = n_ubatch
- No false positives: Valid params don't trigger warnings
- Verified on macOS M3 Pro with embedding model

* Update tools/server/server.cpp

---------

Co-authored-by: ytian218 <ytian218@bloomberg.net>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
---
 tools/server/server.cpp | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/tools/server/server.cpp b/tools/server/server.cpp
index 235ae4e8c0..8538427f73 100644
--- a/tools/server/server.cpp
+++ b/tools/server/server.cpp
@@ -73,8 +73,18 @@ int main(int argc, char ** argv, char ** envp) {
         return 1;
     }
 
+    // validate batch size for embeddings
+    // embeddings require all tokens to be processed in a single ubatch
+    // see https://github.com/ggml-org/llama.cpp/issues/12836
+    if (params.embedding && params.n_batch > params.n_ubatch) {
+        LOG_WRN("%s: embeddings enabled with n_batch (%d) > n_ubatch (%d)\n", __func__, params.n_batch, params.n_ubatch);
+        LOG_WRN("%s: setting n_batch = n_ubatch = %d to avoid assertion failure\n", __func__, params.n_ubatch);
+        params.n_batch = params.n_ubatch;
+    }
+
     if (params.n_parallel < 0) {
         LOG_INF("%s: n_parallel is set to auto, using n_parallel = 4 and kv_unified = true\n", __func__);
+
         params.n_parallel = 4;
         params.kv_unified = true;
     }

From ec98e20021f7611db3bbcf6bb6629fed6e1ce4f0 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Johannes=20G=C3=A4=C3=9Fler?= <johannesg@5d6.de>
Date: Tue, 16 Dec 2025 14:24:00 +0100
Subject: [PATCH 6/6] llama: fix early stop in params_fit if ctx is set
 (#18070)

---
 src/llama.cpp | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)

diff --git a/src/llama.cpp b/src/llama.cpp
index 7ed34b80ae..f69964b6d5 100644
--- a/src/llama.cpp
+++ b/src/llama.cpp
@@ -241,6 +241,13 @@ static void llama_params_fit_impl(
                     global_surplus += memory_reduction;
                     LLAMA_LOG_INFO("%s: context size reduced from %" PRIu32 " to %" PRIu32 " -> need %" PRId64 " MiB less memory in total\n",
                         __func__, hp_nct, cparams->n_ctx, memory_reduction/MiB);
+                    if (global_surplus >= 0) {
+                        if (nd == 1) {
+                            LLAMA_LOG_INFO("%s: entire model can be fit by reducing context\n", __func__);
+                            return;
+                        }
+                        LLAMA_LOG_INFO("%s: entire model should be fit across devices by reducing context\n", __func__);
+                    }
                 } else {
                     LLAMA_LOG_INFO("%s: default model context size is %" PRIu32 " which is <= the min. context size of %" PRIu32 " -> no change\n",
                         __func__, hp_nct, n_ctx_min);
@@ -249,10 +256,6 @@ static void llama_params_fit_impl(
                 LLAMA_LOG_INFO("%s: context size set by user to %" PRIu32 " -> no change\n", __func__, cparams->n_ctx);
             }
         }
-        if (global_surplus >= 0) {
-            LLAMA_LOG_INFO("%s: entire model can be fit across devices by reducing context\n", __func__);
-            return;
-        }
     }
 
     if (mparams->n_gpu_layers != default_mparams.n_gpu_layers) {