root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-27 03:44:35 +00:00
Code Issues Actions 4 Packages Projects Releases Wiki Activity

optimize convert-hf-to-gguf.py for chatglm model

Browse Source 
Signed-off-by: XingXing Qiao <qiaoxx@dingdao.com>
This commit is contained in:
XingXing Qiao 2024-05-16 11:42:53 +08:00
parent f626b7175c
commit f3bc337f43
1 changed files with 8 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
convert-hf-to-gguf.py
View File

@ -2737,13 +2737,15 @@ class ChatGLMModel(Model):
text = piece.encode("utf-8") text = piece.encode("utf-8")
score = 0.0 score = 0.0
if len(piece) != 0 and token_id < 64789: # Referencing the tokenizer Python implementation(https://huggingface.co/THUDM/chatglm3-6b/blob/main/tokenization_chatglm.py),
# it is only valid if it is less than tokenizer.tokenizer.sp_model.vocab_size()
if len(piece) != 0 and token_id < tokenizer.tokenizer.sp_model.vocab_size():
score = tokenizer.tokenizer.sp_model.get_score(token_id) score = tokenizer.tokenizer.sp_model.get_score(token_id)
if len(piece) == 0: if len(piece) == 0:
text = f"[PAD{token_id}]".encode("utf-8") text = f"[PAD{token_id}]".encode("utf-8")
if token_id >= 64789: if token_id >= tokenizer.tokenizer.sp_model.vocab_size():
toktype = SentencePieceTokenTypes.UNKNOWN toktype = SentencePieceTokenTypes.UNKNOWN
tokens.append(text) tokens.append(text)
scores.append(score) scores.append(score)
@ -2773,7 +2775,7 @@ class ChatGLMModel(Model):
special_vocab.add_to_gguf(self.gguf_writer) special_vocab.add_to_gguf(self.gguf_writer)
def set_gguf_parameters(self): def set_gguf_parameters(self):
self.gguf_writer.add_name("ChatGLM-6b-chat") self.gguf_writer.add_name(self.dir_model.name)
n_embed = self.hparams.get("hidden_size", self.hparams.get("n_embed")) n_embed = self.hparams.get("hidden_size", self.hparams.get("n_embed"))
n_head = self.hparams.get("n_head", self.hparams.get("num_attention_heads")) n_head = self.hparams.get("n_head", self.hparams.get("num_attention_heads"))
n_head_kv = self.hparams.get("multi_query_group_num", n_head) n_head_kv = self.hparams.get("multi_query_group_num", n_head)
@ -2789,16 +2791,12 @@ class ChatGLMModel(Model):
self.gguf_writer.add_add_bos_token(False) self.gguf_writer.add_add_bos_token(False)
def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]: def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
del bid # unused
if name.endswith(".rotary_pos_emb.inv_freq"): if name.endswith(".rotary_pos_emb.inv_freq"):
return [] return []
del bid # unused name = name.removeprefix("transformer.")
name = re.sub(r'transformer\.', '', name)
if name == "word_embeddings.weight":
assert self.tensor_names is not None
return [(self.map_tensor_name(name), data_torch)] return [(self.map_tensor_name(name), data_torch)]