From 4e7c8793ae118783cbd390f9dceea8963e019c39 Mon Sep 17 00:00:00 2001 From: ryan-mangeno Date: Thu, 11 Sep 2025 16:41:04 -0400 Subject: [PATCH] fixed asser for equal ubatch seq --- src/llama-graph.cpp | 5 +---- 1 file changed, 1 insertion(+), 4 deletions(-) diff --git a/src/llama-graph.cpp b/src/llama-graph.cpp index 9ca2e579d7..4bd41d0ed6 100644 --- a/src/llama-graph.cpp +++ b/src/llama-graph.cpp @@ -1375,9 +1375,7 @@ ggml_tensor * llm_graph_context::build_attn( // [TAG_NO_CACHE_PAD] // TODO: if ubatch.equal_seqs() == true, we can split the three tensors below into ubatch.n_seqs_unq streams - if (ubatch.n_seqs > 1) { - assert(!ubatch.equal_seqs()); - } + assert(!ubatch.equal_seqs()); ggml_tensor * q = q_cur; ggml_tensor * k = k_cur; @@ -1547,7 +1545,6 @@ ggml_tensor * llm_graph_context::build_attn_with_sinks( // optionally store to KV cache if (k_cur) { const auto & k_idxs = is_swa ? inp->get_k_idxs_swa() : inp->get_k_idxs(); - ggml_build_forward_expand(gf, mctx_cur->cpy_k(ctx0, k_cur, k_idxs, il)); }