root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-31 22:04:35 +00:00
Code Issues Actions 7 Packages Projects Releases Wiki Activity

py : fix position embeddings chop [no ci]

Browse Source
This commit is contained in:
Georgi Gerganov 2024-09-17 13:53:19 +03:00
parent 77723ed69e
commit 49f90de363
No known key found for this signature in database
GPG Key ID: 449E073F9DC10735
1 changed files with 5 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
convert_hf_to_gguf.py
View File

@ -2701,16 +2701,16 @@ class XLMRobertaModel(BertModel):
self.gguf_writer.add_add_eos_token(True)
def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
# position embeddings start at pad_token_id + 1, so just chop down the weight tensor
if name == "embeddings.position_embeddings.weight":
if self._position_offset is not None:
data_torch = data_torch[self._position_offset:,:]
# if name starts with "roberta.", remove the prefix
# e.g. https://huggingface.co/BAAI/bge-reranker-v2-m3/tree/main
if name.startswith("roberta."):
name = name[8:]
# position embeddings start at pad_token_id + 1, so just chop down the weight tensor
if name == "embeddings.position_embeddings.weight":
if self._position_offset is not None:
data_torch = data_torch[self._position_offset:,:]
return super().modify_tensors(data_torch, name, bid)