ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-04-25 08:59:30 +00:00
Code Issues Packages Projects Releases Wiki Activity

qwen3next: avoid extra cont on linear attention output

Browse Source
This commit is contained in:
yurko
2026-02-07 13:30:29 -08:00
parent 0e3891b348
commit 43edfa237b
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/llama-build-context.cpp
View File

@@ -4773,7 +4773,7 @@ ggml_cgraph * llm_build_context::build_qwen3next() {
ggml_tensor * out = llm_build_lora_mm(lctx, ctx0, model.layers[il].ssm_out, final_output);
cb(out, "linear_attn_out", il);
return ggml_cont_2d(ctx0, out, n_embd, n_tok);
return ggml_reshape_2d(ctx0, out, n_embd, n_tok);
};
auto build_layer_attn_linear = [&](ggml_tensor * cur, ggml_tensor * causal_mask, ggml_tensor * identity,