Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

convert : add --fuse-qkv flag to fuse Q/K/V into QKV during HF-to-GGUF conversion (#22780)

Browse Source
This commit is contained in:
PikaPikachu
2026-09-07 00:47:05 +08:00
committed by GitHub
parent 5fdfa62829
commit 465e49b9ce
20 changed files with 396 additions and 125 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-model.cpp
+6
View File
@@ -3236,6 +3236,12 @@ void llama_model_base::create_tensor_qkv(llama_layer & layer, int bid,
layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", bid), {n_embd_, n_embd_qkv}, TENSOR_NOT_REQUIRED | TENSOR_SKIP_IF_VIRTUAL);
if (layer.wqkv) {
layer.wqkv_b = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "bias", bid), {n_embd_qkv}, TENSOR_NOT_REQUIRED | TENSOR_SKIP_IF_VIRTUAL);
// Fused weights may coexist with separate Q/K/V biases in legacy or custom GGUFs.
if (!layer.wqkv_b) {
layer.wq_b = create_tensor(tn(LLM_TENSOR_ATTN_Q, "bias", bid), {n_embd_q_}, TENSOR_NOT_REQUIRED);
layer.wk_b = create_tensor(tn(LLM_TENSOR_ATTN_K, "bias", bid), {n_embd_k_}, TENSOR_NOT_REQUIRED);
layer.wv_b = create_tensor(tn(LLM_TENSOR_ATTN_V, "bias", bid), {n_embd_v_}, TENSOR_NOT_REQUIRED);
}
} else {
layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", bid), {n_embd_, n_embd_q_}, flags);
layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", bid), {n_embd_, n_embd_k_}, flags);