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

support lm_head

Browse source
This commit is contained in:
zhangkaihuo 2024-05-31 16:14:58 +08:00
parent 3b3963c55c
commit 84866fffd2
2 changed files with 4 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

1
gguf-py/gguf/constants.py
View file

@ -631,6 +631,7 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
],
MODEL_ARCH.MINICPM: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT,
MODEL_TENSOR.OUTPUT_NORM,
MODEL_TENSOR.ROPE_FREQS,
MODEL_TENSOR.ATTN_NORM,

5
llama.cpp
View file

@ -4941,7 +4941,8 @@ static bool llm_load_tensors(
// output
{
model.output_norm = ml.create_tensor(ctx_output, tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd});
if (model.arch != LLM_ARCH_MINICPM){
//if (model.arch != LLM_ARCH_MINICPM){
{
model.output = ml.create_tensor(ctx_output_split, tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, false);
// if output is NULL, init from the input tok embed
if (model.output == NULL) {
@ -10120,7 +10121,7 @@ struct llm_build_context {
cb(cur, "lmhead_scaling", -1);
// lm_head
cur = ggml_mul_mat(ctx0, model.tok_embd, cur);
cur = ggml_mul_mat(ctx0, model.output, cur);
cb(cur, "result_output", -1);
ggml_build_forward_expand(gf, cur);