root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-28 12:24:35 +00:00
Code Issues Actions 3 Packages Projects Releases Wiki Activity

resolve comments

Browse Source
This commit is contained in:
Sourab Mangrulkar 2024-03-01 11:09:35 +05:30
parent d62ce1c6b4
commit 10aa6e927e
3 changed files with 24 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
convert-hf-to-gguf.py
View File

@ -98,7 +98,7 @@ class Model:
if (f_rms_eps := self.hparams.get("rms_norm_eps")) is not None: if (f_rms_eps := self.hparams.get("rms_norm_eps")) is not None:
self.gguf_writer.add_layer_norm_rms_eps(f_rms_eps) self.gguf_writer.add_layer_norm_rms_eps(f_rms_eps)
if (f_norm_eps := self.find_hparam(["layer_norm_eps", "layer_norm_epsilon"], optional=True)) is not None: if (f_norm_eps := self.find_hparam(["layer_norm_eps", "layer_norm_epsilon", "norm_epsilon"], optional=True)) is not None:
self.gguf_writer.add_layer_norm_eps(f_norm_eps) self.gguf_writer.add_layer_norm_eps(f_norm_eps)
if (n_experts := self.hparams.get("num_local_experts")) is not None: if (n_experts := self.hparams.get("num_local_experts")) is not None:
self.gguf_writer.add_expert_count(n_experts) self.gguf_writer.add_expert_count(n_experts)
@ -220,6 +220,8 @@ class Model:
return NomicBertModel return NomicBertModel
if model_architecture == "GemmaForCausalLM": if model_architecture == "GemmaForCausalLM":
return GemmaModel return GemmaModel
if model_architecture == "Starcoder2ForCausalLM":
return StarCoderModel2
return Model return Model
def _is_model_safetensors(self) -> bool: def _is_model_safetensors(self) -> bool:
@ -926,6 +928,10 @@ class StarCoderModel(Model):
self.gguf_writer.add_layer_norm_eps(self.hparams["layer_norm_epsilon"]) self.gguf_writer.add_layer_norm_eps(self.hparams["layer_norm_epsilon"])
self.gguf_writer.add_file_type(self.ftype) self.gguf_writer.add_file_type(self.ftype)
class StarCoderModel2(Model):
def set_vocab(self):
self._set_vocab_gpt2()
class RefactModel(Model): class RefactModel(Model):
def set_gguf_parameters(self): def set_gguf_parameters(self):

30
gguf-py/gguf/constants.py
View File

@ -275,21 +275,6 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.FFN_DOWN, MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP, MODEL_TENSOR.FFN_UP,
], ],
MODEL_ARCH.STARCODER2: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,
MODEL_TENSOR.OUTPUT,
MODEL_TENSOR.ROPE_FREQS,
MODEL_TENSOR.ATTN_NORM,
MODEL_TENSOR.ATTN_Q,
MODEL_TENSOR.ATTN_K,
MODEL_TENSOR.ATTN_V,
MODEL_TENSOR.ATTN_OUT,
MODEL_TENSOR.ATTN_ROT_EMBD,
MODEL_TENSOR.FFN_NORM,
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
],
MODEL_ARCH.BERT: [ MODEL_ARCH.BERT: [
MODEL_TENSOR.TOKEN_EMBD, MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.TOKEN_EMBD_NORM, MODEL_TENSOR.TOKEN_EMBD_NORM,
@ -543,6 +528,21 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.FFN_UP, MODEL_TENSOR.FFN_UP,
MODEL_TENSOR.FFN_NORM, MODEL_TENSOR.FFN_NORM,
], ],
MODEL_ARCH.STARCODER2: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,
MODEL_TENSOR.OUTPUT,
MODEL_TENSOR.ROPE_FREQS,
MODEL_TENSOR.ATTN_NORM,
MODEL_TENSOR.ATTN_Q,
MODEL_TENSOR.ATTN_K,
MODEL_TENSOR.ATTN_V,
MODEL_TENSOR.ATTN_OUT,
MODEL_TENSOR.ATTN_ROT_EMBD,
MODEL_TENSOR.FFN_NORM,
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
],
# TODO # TODO
} }

4
llama.cpp
View File

@ -211,8 +211,8 @@ enum llm_arch {
LLM_ARCH_INTERNLM2, LLM_ARCH_INTERNLM2,
LLM_ARCH_MINICPM, LLM_ARCH_MINICPM,
LLM_ARCH_GEMMA, LLM_ARCH_GEMMA,
LLM_ARCH_STARCODER2,
LLM_ARCH_UNKNOWN, LLM_ARCH_UNKNOWN,
LLM_ARCH_STARCODER2
}; };
static std::map<llm_arch, const char *> LLM_ARCH_NAMES = { static std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
@ -7735,7 +7735,7 @@ struct llm_build_context {
cur = llm_build_norm(ctx0, cur, hparams, cur = llm_build_norm(ctx0, cur, hparams,
model.output_norm, model.output_norm_b, model.output_norm, model.output_norm_b,
LLM_NORM_RMS, cb, -1); LLM_NORM, cb, -1);
cb(cur, "result_norm", -1); cb(cur, "result_norm", -1);
// lm_head // lm_head