From 1099cbf694a8d5d85b6ebd0852c21b53bad2ccce Mon Sep 17 00:00:00 2001 From: Yee Man Chan Date: Wed, 7 Jan 2026 18:42:31 +0800 Subject: [PATCH] build_kda_autoregressive is implemented to replace build_kda_recurrent for faster inference. sync'd to b7682 --- src/models/kimi-linear.cpp | 355 +++++++++++-------------------------- src/models/models.h | 9 +- 2 files changed, 110 insertions(+), 254 deletions(-) diff --git a/src/models/kimi-linear.cpp b/src/models/kimi-linear.cpp index 013926e544..270f9e6e6b 100644 --- a/src/models/kimi-linear.cpp +++ b/src/models/kimi-linear.cpp @@ -20,14 +20,16 @@ llm_build_kimi_linear::llm_build_kimi_linear(const llama_model & model, const ll // Output ids for selecting which tokens to output 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_tensor * chunked_causal_mask = + 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 * chunked_identity = ggml_diag(ctx0, ggml_fill_inplace(ctx0, ggml_new_tensor_1d(ctx0, GGML_TYPE_F32, CHUNK_SIZE), 1.0f)); + ggml_tensor * chunked_diag_mask = ggml_add(ctx0, chunked_causal_mask, chunked_identity); - ggml_build_forward_expand(gf, causal_mask); - ggml_build_forward_expand(gf, identity); + ggml_build_forward_expand(gf, chunked_causal_mask); + ggml_build_forward_expand(gf, chunked_identity); + ggml_build_forward_expand(gf, chunked_diag_mask); // Kimi dimension constants const int64_t n_head = hparams.n_head(); @@ -263,9 +265,9 @@ llm_build_kimi_linear::llm_build_kimi_linear(const llama_model & model, const ll ggml_tensor * state = build_rs(inp_rs, ssm_states_all, hparams.n_embd_s(), n_seqs); state = ggml_reshape_4d(ctx0, state, head_dim, head_dim, n_head, n_seqs); // Choose between build_kda_chunking and build_kda_recurrent based on n_tokens - ggml_tensor * attn_out = n_seq_tokens > CHUNK_SIZE ? - build_kda_chunking(Qcur, Kcur, Vcur, g1, beta, state, causal_mask, identity, il) : - build_kda_recurrent(Qcur, Kcur, Vcur, g1, beta, state, causal_mask, identity, il); + ggml_tensor * attn_out = n_seq_tokens == 1 ? + build_kda_autoregressive(Qcur, Kcur, Vcur, g1, beta, state, il) : + build_kda_chunking(Qcur, Kcur, Vcur, g1, beta, state, chunked_causal_mask, chunked_identity, chunked_diag_mask, il); cb(attn_out, "attn_out", il); // The tensors were concatenated 1d, so we need to extract them 1d as well @@ -464,6 +466,7 @@ ggml_tensor * llm_build_kimi_linear::build_kda_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)); @@ -519,8 +522,6 @@ ggml_tensor * llm_build_kimi_linear::build_kda_chunking( 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); - ggml_tensor * causal_diag_mask = ggml_add(ctx0, causal_mask, identity); - cb(q, "q_perm", il); cb(k, "k_perm", il); cb(v, "v_perm", il); @@ -557,21 +558,6 @@ ggml_tensor * llm_build_kimi_linear::build_kda_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); - - 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); - const int64_t HB = H_k * n_seqs; q = ggml_cont_4d(ctx0, q, S_k, chunk_size, n_chunks, HB); @@ -588,6 +574,14 @@ ggml_tensor * llm_build_kimi_linear::build_kda_chunking( ggml_tensor * gk_cumsum = ggml_cumsum(ctx0, gk); cb(gk_cumsum, "gk_cumsum", il); +/* + https://github.com/fla-org/flash-linear-attention/blob/main/fla/ops/kda/naive.py + + for i in range(T): + k_i = k[..., i, :] + g_i = g[..., i:i+1, :] + A[..., i] = torch.einsum('... c d, ... d -> ... c', k * (g - g_i).exp(), k_i) +*/ const int64_t CHB = n_chunks * H_v * n_seqs; ggml_tensor * g_i = ggml_reshape_4d(ctx0, gk_cumsum, chunk_size, 1, S_k, CHB); @@ -599,9 +593,9 @@ ggml_tensor * llm_build_kimi_linear::build_kda_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); cb(decay_mask, "decay_mask_exp", il); // k [S,BT,NT,H*B] k_per [BT,S,NT,H*B] @@ -620,19 +614,27 @@ ggml_tensor * llm_build_kimi_linear::build_kda_chunking( Akk = ggml_reshape_4d(ctx0, Akk, chunk_size, chunk_size, n_chunks, H_k * n_seqs); Akk = ggml_mul(ctx0, Akk, beta); - Akk = ggml_neg(ctx0, ggml_mul(ctx0, Akk, chunked_mask)); + Akk = ggml_neg(ctx0, ggml_mul(ctx0, Akk, causal_mask)); cb(Akk, "attn_pre_solve", il); - ggml_tensor * attn_lower = ggml_mul(ctx0, Akk, chunked_mask); - ggml_tensor * lhs = ggml_sub(ctx0, ggml_repeat(ctx0, chunked_identity, attn_lower), attn_lower); + // 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, Akk, 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, Akk, true, true, false); - Akk = ggml_mul(ctx0, lin_solve, chunked_mask); - Akk = ggml_add(ctx0, Akk, chunked_identity); + Akk = ggml_mul(ctx0, lin_solve, causal_mask); + Akk = ggml_add(ctx0, Akk, identity); cb(Akk, "attn_solved", il); + // u = (A*beta[..., None, :]) @ v aka U_[t] ggml_tensor * vb = ggml_mul_mat(ctx0, ggml_cont(ctx0, ggml_transpose(ctx0, v_beta)), Akk); gk_cumsum = ggml_cont_4d(ctx0, ggml_permute(ctx0, gk_cumsum, 1, 0, 2, 3), S_k, chunk_size, n_chunks, HB); @@ -650,7 +652,6 @@ ggml_tensor * llm_build_kimi_linear::build_kda_chunking( cb(new_state, "new_state", il); for (int64_t chunk = 0; chunk < n_chunks; chunk++) { -// for (int64_t chunk = 0; chunk < 1; chunk++) { // extract one chunk worth of data auto chunkify = [=](ggml_tensor * t) { return ggml_cont(ctx0, ggml_view_4d(ctx0, t, t->ne[0], chunk_size, 1, t->ne[3], @@ -672,15 +673,22 @@ ggml_tensor * llm_build_kimi_linear::build_kda_chunking( ggml_tensor * gk_cs_chunk_j_bc = ggml_repeat_4d(ctx0, gk_cs_chunk_j, chunk_size, chunk_size, S_k, HB); ggml_tensor * decay_mask_chunk = ggml_sub(ctx0, gk_cs_chunk_j_bc, gk_cs_chunk_i); cb(decay_mask_chunk, "decay_mask_chunk", il); - decay_mask_chunk = ggml_mul(ctx0, decay_mask_chunk, chunked_diag_mask); + decay_mask_chunk = ggml_mul(ctx0, decay_mask_chunk, diag_mask); decay_mask_chunk = ggml_exp(ctx0, decay_mask_chunk); - decay_mask_chunk = ggml_mul(ctx0, decay_mask_chunk, chunked_diag_mask); + decay_mask_chunk = ggml_mul(ctx0, decay_mask_chunk, diag_mask); cb(decay_mask_chunk, "decay_mask_chunk_exp", il); ggml_tensor * k_cumdecay_chunk = chunkify(k_cumdecay); ggml_tensor * gkexp_chunk = ggml_exp(ctx0, gk_cs_chunk); +/* + https://github.com/fla-org/flash-linear-attention/blob/main/fla/ops/kda/naive.py + for j in range(BT): + k_j = k[:, :, i, j] + g_j = g[:, :, i, j:j+1, :] + A[..., j] = torch.einsum('... c d, ... d -> ... c', q_i * (g_i - g_j).exp(), k_j) +*/ ggml_tensor * k_chunk_i = ggml_cont(ctx0, ggml_permute(ctx0, k_chunk, 2, 0, 1, 3)); ggml_tensor * k_chunk_i_bc = ggml_repeat_4d(ctx0, k_chunk_i, chunk_size, chunk_size, S_k, HB); ggml_tensor * q_chunk_j = ggml_cont(ctx0, ggml_permute(ctx0, q_chunk, 2, 1, 0, 3)); @@ -689,7 +697,7 @@ ggml_tensor * llm_build_kimi_linear::build_kda_chunking( kq = ggml_mul(ctx0, kq, k_chunk_i_bc); ggml_tensor * Aqk = ggml_mul(ctx0, kq, decay_mask_chunk); - Aqk = ggml_mul(ctx0, Aqk, ggml_add(ctx0, chunked_identity, chunked_mask)); + Aqk = ggml_mul(ctx0, Aqk, ggml_add(ctx0, identity, causal_mask)); Aqk = ggml_cont(ctx0, ggml_permute(ctx0, Aqk, 1, 2, 0, 3)); Aqk = ggml_sum_rows(ctx0, Aqk); Aqk = ggml_scale(ctx0, Aqk, scale); // scale q @@ -697,20 +705,26 @@ ggml_tensor * llm_build_kimi_linear::build_kda_chunking( 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); -// new_state [S,S,1,H*B] k_cumdecay_chunk [S,BT,1,H*B] + // new_state [S,S,1,H*B] k_cumdecay_chunk [S,BT,1,H*B] + // v_prime = (k_cumdecay[:, :, i]) @ last_recurrent_state or W_[t] @ S_[t] ggml_tensor * v_prime = ggml_mul_mat(ctx0, state_t, k_cumdecay_chunk); + // v_new = v_i - v_prime or U_[t] - W_[t]*S_[t] ggml_tensor * v_new = ggml_sub(ctx0, ggml_repeat(ctx0, vb_chunk, v_prime), v_prime); ggml_tensor * v_new_t = ggml_cont(ctx0, ggml_transpose(ctx0, v_new)); -// q_chunk [S,BT,1,H*B] gkexp_chunk [S,BT,1,H*B] + // q_chunk [S,BT,1,H*B] gkexp_chunk [S,BT,1,H*B] + // attn_inter = (q_i * g[:, :, i, :, None].exp()) @ last_recurrent_state + // or Gamma_[t]*Q_]t] @ S ggml_tensor * q_gk_exp = ggml_mul(ctx0, q_chunk, gkexp_chunk); ggml_tensor * attn_inter = ggml_mul_mat(ctx0, state_t, q_gk_exp); attn_inter = ggml_scale(ctx0, attn_inter, scale); // scale q -// v_new_t [S,BT,1,H*B] Aqk [BT,BT,1,H*B] + // v_new_t [S,BT,1,H*B] Aqk [BT,BT,1,H*B] + // core_attn_out[:, :, i] = attn_inter + attn @ v_new or A' @ (U_[t] - W_[t]*S_[t]) ggml_tensor * v_attn = ggml_mul_mat(ctx0, v_new_t, Aqk); + // o[:, :, i] = (q_i * g_i.exp()) @ S + A @ v_i ggml_tensor * core_attn_out_chunk = ggml_add(ctx0, attn_inter, v_attn); core_attn_out = core_attn_out == nullptr ? core_attn_out_chunk : ggml_concat(ctx0, core_attn_out, core_attn_out_chunk, 1); @@ -728,6 +742,7 @@ ggml_tensor * llm_build_kimi_linear::build_kda_chunking( ggml_tensor * key_gkdiff = ggml_mul(ctx0, k_chunk, gk_diff_exp); + // rearrange((g_i[:,:,-1:] - g_i).exp()*k_i, 'b h c k -> b h k c') @ (U_[t] - W_[t] @ S) ggml_tensor * kgdmulvnew = ggml_mul_mat(ctx0, v_new_t, ggml_cont(ctx0, ggml_transpose(ctx0, key_gkdiff))); new_state = ggml_add(ctx0, @@ -750,256 +765,98 @@ ggml_tensor * llm_build_kimi_linear::build_kda_chunking( return ggml_concat(ctx0, flat_output, flat_state, 0); } -ggml_tensor * llm_build_kimi_linear::build_kda_recurrent( +ggml_tensor * llm_build_kimi_linear::build_kda_autoregressive( ggml_tensor * q, ggml_tensor * k, ggml_tensor * v, ggml_tensor * gk, ggml_tensor * beta, - ggml_tensor * state, - ggml_tensor * causal_mask, - ggml_tensor * identity, + ggml_tensor * state, 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(v)); GGML_ASSERT(ggml_is_contiguous(gk)); 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]; const int64_t n_seqs = q->ne[3]; - + const int64_t S_v = v->ne[0]; const int64_t H_v = v->ne[1]; - + + GGML_ASSERT(n_tokens == 1); GGML_ASSERT(v->ne[2] == n_tokens); GGML_ASSERT(k->ne[2] == n_tokens); - GGML_ASSERT(gk->ne[0] == S_k && gk->ne[1] == H_v && gk->ne[2] == n_tokens && gk->ne[3] == n_seqs); + GGML_ASSERT(gk->ne[0] == S_k && gk->ne[1] == H_k && gk->ne[2] == n_tokens && gk->ne[3] == n_seqs); GGML_ASSERT(beta->ne[0] == H_v && beta->ne[2] == n_tokens && beta->ne[3] == n_seqs); - GGML_ASSERT(state->ne[0] == S_v && state->ne[1] == S_v && state->ne[2] == H_v && state->ne[3] == n_seqs); - + GGML_ASSERT(state->ne[0] == S_v && state->ne[1] == S_k && state->ne[2] == H_v && state->ne[3] == n_seqs); + GGML_ASSERT(q->ne[0] == S_k && q->ne[1] == H_k && q->ne[2] == n_tokens && q->ne[3] == n_seqs); GGML_ASSERT(k->ne[0] == S_k && k->ne[1] == H_k && k->ne[2] == n_tokens && k->ne[3] == n_seqs); - - 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; - - 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); - } - - const float scale = 1.0f / sqrtf(S_v); - - beta = ggml_sigmoid(ctx0, beta); - ggml_tensor * causal_diag_mask = ggml_add(ctx0, causal_mask, identity); + GGML_ASSERT(H_k == H_v); // we did a repeat to make sure this is the case + + 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); + + const float scale = 1.0f / sqrtf(S_v); + + q = ggml_scale(ctx0, q, scale); + beta = ggml_sigmoid(ctx0, beta); cb(q, "q_in", il); cb(k, "k_in", il); cb(v, "v_in", il); cb(beta, "beta_in", il); cb(gk, "gk_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); - gk = ggml_cont_4d(ctx0, ggml_permute(ctx0, gk, 1, 2, 0, 3), n_tokens, S_k, 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(gk, "gk_perm", il); - cb(state, "state_in", il); - - 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); - // ========================================================================= - // Compute cumulative sum of gk per key dimension - // gk_cumsum: [S_k, n_tokens, H_k, n_seqs] - cumsum along dim 1 (tokens) - // ========================================================================= - ggml_tensor * gk_cumsum = ggml_cumsum(ctx0, gk); - cb(gk_cumsum, "gk_cumsum", il); +// g [H,1,B,1] g_t [1,H,B,1] => [1,1,H,B] +// gk [S,H,1,B] => [S,1,H,B] gk_t [1,S,H,B] +// beta [H,1,1,B] beta_t [1,H,1,B] => [1,1,H,B] + gk = ggml_reshape_4d(ctx0, gk, S_k, 1, H_k, n_seqs); + ggml_tensor * gk_t = ggml_cont(ctx0, ggml_transpose(ctx0, gk)); + ggml_tensor * beta_t = ggml_reshape_4d(ctx0, ggml_transpose(ctx0, beta), 1, 1, H_k, n_seqs); - // Scale k and k_beta - - ggml_tensor * k_beta = ggml_mul(ctx0, k, beta); - ggml_tensor * v_beta = ggml_mul(ctx0, v, beta); - - cb(k_beta, "k_beta", il); - cb(v_beta, "v_beta", il); + // Apply exponential to gk_t + gk_t = ggml_exp(ctx0, gk_t); + // Apply the gated delta rule for the single timestep + // last_recurrent_state = last_recurrent_state * gk_t + // S = S * g_i[..., None].exp() + state = ggml_mul(ctx0, state, gk_t); - -/* - https://github.com/fla-org/flash-linear-attention/blob/main/fla/ops/kda/naive.py - - for i in range(T): - k_i = k[..., i, :] - g_i = g[..., i:i+1, :] - A[..., i] = torch.einsum('... c d, ... d -> ... c', k * (g - g_i).exp(), k_i) -*/ - const int64_t HB = H_k * n_seqs; - ggml_tensor * k_per = ggml_cont(ctx0, ggml_permute(ctx0, k, 1, 0, 2, 3)); - ggml_tensor * k_i = ggml_reshape_4d(ctx0, k_per, n_tokens, 1, S_k, HB); - ggml_tensor * k_i_bc = ggml_repeat_4d(ctx0, k_i, n_tokens, n_tokens, S_k, HB); - ggml_tensor * g_i = ggml_reshape_4d(ctx0, gk_cumsum, n_tokens, 1, S_k, HB); - ggml_tensor * g_i_bc = ggml_repeat_4d(ctx0, g_i, n_tokens, n_tokens, S_k, HB); // [S_k, chunk_size, 1, HB] -> [S_k, chunk_size, chunk_size, HB] - - ggml_tensor * k_j = ggml_reshape_4d(ctx0, k_per, 1, n_tokens, S_k, HB); - ggml_tensor * k_j_bc = ggml_repeat_4d(ctx0, k_j, n_tokens, n_tokens, S_k, HB); - - ggml_tensor * g_j = ggml_reshape_4d(ctx0, gk_cumsum, 1, n_tokens, S_k, HB); - ggml_tensor * g_j_bc = ggml_repeat_4d(ctx0, g_j, n_tokens, n_tokens, S_k, HB); // [S_k, 1, chunk_size, HB] -> [S_k, chunk_size, chunk_size, HB] - - ggml_tensor * decay_mask = ggml_sub(ctx0, g_j_bc, g_i_bc); - cb(decay_mask, "decay_mask", il); - decay_mask = ggml_mul(ctx0, decay_mask, causal_diag_mask); - decay_mask = ggml_exp(ctx0, decay_mask); - decay_mask = ggml_mul(ctx0, decay_mask, causal_diag_mask); - cb(decay_mask, "decay_mask_exp", il); - - ggml_tensor * Akk = ggml_mul(ctx0, decay_mask, k_j_bc); - Akk = ggml_mul(ctx0, Akk, k_i_bc); - - Akk = ggml_cont(ctx0, ggml_permute(ctx0, Akk, 1, 2, 0, 3)); - Akk = ggml_sum_rows(ctx0, Akk); - - Akk = ggml_reshape_4d(ctx0, Akk, n_tokens, n_tokens, H_k, n_seqs); - - Akk = ggml_mul(ctx0, Akk, beta); - Akk = ggml_neg(ctx0, ggml_mul(ctx0, Akk, causal_mask)); - - cb(Akk, "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, Akk, 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, Akk, true, true, false); - Akk = ggml_mul(ctx0, lin_solve, causal_mask); - Akk = ggml_add(ctx0, Akk, identity); - - gk_cumsum = ggml_cont(ctx0, ggml_permute(ctx0, gk_cumsum, 1, 0, 2, 3)); // back to [S_k, n_tokens, H_k, n_seqs] - - // u = (A*beta[..., None, :]) @ v aka U_[t] - ggml_tensor * vb = ggml_mul_mat(ctx0, ggml_cont(ctx0, ggml_transpose(ctx0, v_beta)), Akk); - cb(vb, "value_beta", il); - - // k_cumdecay = attn @ (k_beta * g.exp().unsqueeze(-1)) or W_[t] - ggml_tensor * gkexp = ggml_exp(ctx0, gk_cumsum); // [S,T,H,B] - - ggml_tensor * kbeta_gkexp = ggml_mul(ctx0, k_beta, gkexp); - cb(kbeta_gkexp, "kbeta_gkexp", il); - - ggml_tensor * k_cumdecay = ggml_mul_mat(ctx0, ggml_cont(ctx0, ggml_transpose(ctx0, kbeta_gkexp)), Akk); - cb(k_cumdecay, "k_cumdecay", il); - -/* - https://github.com/fla-org/flash-linear-attention/blob/main/fla/ops/kda/naive.py - - for j in range(BT): - k_j = k[:, :, i, j] - g_j = g[:, :, i, j:j+1, :] - A[..., j] = torch.einsum('... c d, ... d -> ... c', q_i * (g_i - g_j).exp(), k_j) -*/ - ggml_tensor * q_per = ggml_cont(ctx0, ggml_permute(ctx0, q, 1, 0, 2, 3)); - ggml_tensor * q_j = ggml_reshape_4d(ctx0, q_per, 1, n_tokens, S_k, HB); - ggml_tensor * q_j_bc = ggml_repeat_4d(ctx0, q_j, n_tokens, n_tokens, S_k, HB); - ggml_tensor * kq = ggml_mul(ctx0, decay_mask, q_j_bc); - kq = ggml_mul(ctx0, kq, k_i_bc); - kq = ggml_cont(ctx0, ggml_permute(ctx0, kq, 1, 2, 0, 3)); - - ggml_tensor * Aqk = ggml_sum_rows(ctx0, kq); - Aqk = ggml_cont(ctx0, ggml_reshape_4d(ctx0, Aqk, n_tokens, n_tokens, H_k, n_seqs)); - Aqk = ggml_mul(ctx0, Aqk, ggml_add(ctx0, identity, causal_mask)); - Aqk = ggml_scale(ctx0, Aqk, scale); // scale q - cb(Aqk, "attn_decay_key", il); - ggml_tensor * state_t = ggml_cont(ctx0, ggml_transpose(ctx0, state)); - - // v_prime = (k_cumdecay[:, :, i]) @ last_recurrent_state or W_[t] @ S_[t] - ggml_tensor * v_prime = ggml_mul_mat(ctx0, state_t, k_cumdecay); - cb(v_prime, "v_prime", il); +// state [S,S,H,B] k [S,1,H,B] k_state [S_v,1,H,B] + k = ggml_reshape_4d(ctx0, k, S_k, 1, H_k, n_seqs); + ggml_tensor * k_state = ggml_mul_mat(ctx0, state_t, k); - // v_new = v_i - v_prime or U_[t] - W_[t]*S_[t] - ggml_tensor * v_new = ggml_sub(ctx0, ggml_repeat(ctx0, vb, v_prime), v_prime); + // v_i - (k_i[..., None] * S).sum(-2) + v = ggml_reshape_4d(ctx0, v, S_v, 1, H_v, n_seqs); + ggml_tensor * v_diff = ggml_sub(ctx0, v, k_state); - // v_new_t [T.S.H,B] - ggml_tensor * v_new_t = ggml_cont(ctx0, ggml_transpose(ctx0, v_new)); + // b_i[..., None] * k_i + ggml_tensor * k_beta = ggml_mul(ctx0, k, beta_t); - cb(v_new, "v_new", il); + // S = S + torch.einsum('b h k, b h v -> b h k v', b_i[..., None] * k_i, v_i - (k_i[..., None] * S).sum(-2)) + // v_diff_t [1,S_v,H,B] k_beta_t [1,S_k,H,B] state [S_v,S_k,H,B] + state = ggml_add(ctx0, state, ggml_mul_mat(ctx0, ggml_cont(ctx0, ggml_transpose(ctx0, v_diff)), ggml_cont(ctx0, ggml_transpose(ctx0, k_beta)))); - // attn_inter = (q_i * g[:, :, i, :, None].exp()) @ last_recurrent_state - // or Gamma_[t]*Q_]t] @ S - ggml_tensor * q_gk_exp = ggml_mul(ctx0, q, gkexp); - ggml_tensor * attn_inter = ggml_mul_mat(ctx0, state_t, q_gk_exp); - // scale q at attn_inter as suggested in chunk_gla_fwd_kernel_o of - // github.com/fla-org/flash-linear-attention/fla/ops/gla/chunk.py - attn_inter = ggml_scale(ctx0, attn_inter, scale); // scale q - - cb(attn_inter, "attn_inter", il); - - // core_attn_out[:, :, i] = attn_inter + attn @ v_new or A' @ (U_[t] - W_[t]*S_[t]) - ggml_tensor * v_attn = ggml_mul_mat(ctx0, v_new_t, Aqk); - - cb(v_attn, "v_attn", il); - - // o[:, :, i] = (q_i * g_i.exp()) @ S + A @ v_i - ggml_tensor * core_attn_out = ggml_add(ctx0, attn_inter, v_attn); - - cb(core_attn_out, "core_attn_out", il); - - ggml_tensor * gk_cum_last = - ggml_cont(ctx0, ggml_view_4d(ctx0, gk_cumsum, gk_cumsum->ne[0], 1, gk_cumsum->ne[2], gk_cumsum->ne[3], - gk_cumsum->nb[1], gk_cumsum->nb[2], gk_cumsum->nb[3], - gk_cumsum->nb[1] * (gk_cumsum->ne[1] - 1))); - cb(gk_cum_last, "gk_cum_last", il); - - ggml_tensor * gkexp_last = ggml_exp(ctx0, ggml_cont(ctx0, ggml_transpose(ctx0, gk_cum_last))); - cb(gkexp_last, "gkexp_last", il); - - ggml_tensor * gk_diff = ggml_neg(ctx0, ggml_sub(ctx0, gk_cumsum, gk_cum_last)); - cb(gk_diff, "gk_diff", il); - - ggml_tensor * gk_diff_exp = ggml_exp(ctx0, gk_diff); - cb(gk_diff_exp, "gk_diff_exp", il); - - ggml_tensor * key_gkdiff = ggml_mul(ctx0, k, gk_diff_exp); - cb(key_gkdiff, "key_gkdiff", il); - - // rearrange((g_i[:,:,-1:] - g_i).exp()*k_i, 'b h c k -> b h k c') @ (U_[t] - W_[t] @ S) - ggml_tensor * kgkdmulvnew = ggml_mul_mat(ctx0, v_new_t, ggml_cont(ctx0, ggml_transpose(ctx0, key_gkdiff))); - cb(kgkdmulvnew, "kgkdmulvnew", il); - - state = ggml_add(ctx0, ggml_mul(ctx0, state, gkexp_last), kgkdmulvnew); + q = ggml_reshape_4d(ctx0, q, S_k, 1, H_k, n_seqs); + state_t = ggml_cont(ctx0, ggml_transpose(ctx0, state)); + ggml_tensor * core_attn_out = ggml_mul_mat(ctx0, state_t, 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); } diff --git a/src/models/models.h b/src/models/models.h index ba2b905c5e..3ed00aae32 100644 --- a/src/models/models.h +++ b/src/models/models.h @@ -288,26 +288,25 @@ struct llm_build_kimi_linear : public llm_graph_context_mamba { llm_build_kimi_linear(const llama_model & model, const llm_graph_params & params); private: const llama_model & model; - ggml_tensor * build_kda_recurrent( + ggml_tensor * build_kda_autoregressive( ggml_tensor * q, ggml_tensor * k, ggml_tensor * v, - ggml_tensor * g, + ggml_tensor * gk, ggml_tensor * beta, ggml_tensor * state, - ggml_tensor * causal_mask, - ggml_tensor * identity, int il); ggml_tensor * build_kda_chunking( ggml_tensor * q, ggml_tensor * k, ggml_tensor * v, - ggml_tensor * g, + ggml_tensor * gk, ggml_tensor * beta, ggml_tensor * state, ggml_tensor * causal_mask, ggml_tensor * identity, + ggml_tensor * diag_mask, int il); };