CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-07 09:57:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix

Browse Source
This commit is contained in:
younesbelkada
2025-07-03 15:10:18 +04:00
parent 71a6848e2d
commit 03568c9358
1 changed files with 1 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/llama-model.cpp
View File

@@ -14674,8 +14674,6 @@ struct llm_build_falcon_h1 : public llm_graph_context {
// Build the inputs in the recurrent & kv cache // Build the inputs in the recurrent & kv cache
auto * inp = build_inp_mem_hybrid(); auto * inp = build_inp_mem_hybrid();
auto * inp_attn = build_attn_inp_kv_unified();
const float kq_scale = hparams.f_attention_scale == 0.0f ? 1.0f/sqrtf(float(n_embd_head)) : hparams.f_attention_scale; const float kq_scale = hparams.f_attention_scale == 0.0f ? 1.0f/sqrtf(float(n_embd_head)) : hparams.f_attention_scale;
for (int il = 0; il < n_layer; ++il) { for (int il = 0; il < n_layer; ++il) {
@@ -14718,7 +14716,7 @@ struct llm_build_falcon_h1 : public llm_graph_context {
cb(Kcur, "Kcur", il); cb(Kcur, "Kcur", il);
cb(Vcur, "Vcur", il); cb(Vcur, "Vcur", il);
ggml_tensor * attn_out = build_attn(inp_attn, gf, ggml_tensor * attn_out = build_attn(inp, gf,
model.layers[il].wo, NULL, model.layers[il].wo, NULL,
Qcur, Kcur, Vcur, nullptr, nullptr, kq_scale, il); Qcur, Kcur, Vcur, nullptr, nullptr, kq_scale, il);
attn_out = ggml_scale(ctx0, attn_out, hparams.attention_out_multiplier); attn_out = ggml_scale(ctx0, attn_out, hparams.attention_out_multiplier);