vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

fix format

Browse source
This commit is contained in:
liyuhang 2025-02-01 08:28:20 +08:00
parent 5f13c244b3
commit 77b0b28d3e
2 changed files with 6 additions and 12 deletions
Download patch file Download diff file Expand all files Collapse all files

6
src/llama-model.cpp
View file

@ -1094,14 +1094,14 @@ void llama_model::load_hparams(llama_model_loader & ml) {
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
switch (hparams.n_layer) { switch (hparams.n_layer) {
case 28: { case 28: {
if (hparams.n_head(0)==16) { if (hparams.n_head(0) == 16) {
type = LLM_TYPE_1_5B; type = LLM_TYPE_1_5B;
} else { } else {
type = LLM_TYPE_6B; type = LLM_TYPE_6B;
} }
} break; } break;
case 40: { case 40: {
if(hparams.n_head(0)==24){ if(hparams.n_head(0) == 24){
type = LLM_TYPE_4B; type = LLM_TYPE_4B;
} else { } else {
type = LLM_TYPE_9B; type = LLM_TYPE_9B;
@ -3083,7 +3083,7 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED); layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED);
layer.bqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "bias", i), {n_embd + 2*n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED); layer.bqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "bias", i), {n_embd + 2*n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED);
if(layer.wqkv == nullptr) { if (layer.wqkv == nullptr) {
layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd_head_k * n_head}, 0); layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd_head_k * n_head}, 0);
layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_k_gqa}, 0); layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_k_gqa}, 0);
layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_v_gqa}, 0); layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_v_gqa}, 0);

12
src/llama.cpp
View file

@ -7217,22 +7217,16 @@ struct llm_build_context {
struct ggml_tensor * Vcur = nullptr; struct ggml_tensor * Vcur = nullptr;
if (model.type == LLM_TYPE_1_5B || model.type == LLM_TYPE_4B || model.type == LLM_TYPE_9B) { if (model.type == LLM_TYPE_1_5B || model.type == LLM_TYPE_4B || model.type == LLM_TYPE_9B) {
Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur); Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
cb(Qcur, "Qcur", il);
if (model.layers[il].bq) { if (model.layers[il].bq) {
Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq); Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
cb(Qcur, "Qcur", il);
} }
Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur); Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
cb(Kcur, "Kcur", il);
if (model.layers[il].bk) { if (model.layers[il].bk) {
Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk); Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
cb(Kcur, "Kcur", il);
} }
Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur); Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
cb(Vcur, "Vcur", il);
if (model.layers[il].bv) { if (model.layers[il].bv) {
Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv); Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
cb(Vcur, "Vcur", il);
} }
} else { } else {
cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur); cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur);
@ -7244,10 +7238,10 @@ struct llm_build_context {
Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd, n_tokens, cur->nb[1], 0*sizeof(float)*(n_embd))); Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd, n_tokens, cur->nb[1], 0*sizeof(float)*(n_embd)));
Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd))); Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd)));
Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa))); Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa)));
cb(Qcur, "Qcur", il);
cb(Kcur, "Kcur", il);
cb(Vcur, "Vcur", il);
} }
cb(Qcur, "Qcur", il);
cb(Kcur, "Kcur", il);
cb(Vcur, "Vcur", il);
//printf("freq_base: %f freq_scale: %f ext_factor: %f attn_factor: %f\n", freq_base, freq_scale, ext_factor, attn_factor); //printf("freq_base: %f freq_scale: %f ext_factor: %f attn_factor: %f\n", freq_base, freq_scale, ext_factor, attn_factor);
Qcur = ggml_rope_ext( Qcur = ggml_rope_ext(
ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, nullptr, ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, nullptr,