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

convert : add support for Roberta embeddings (#10695)

Browse source
This commit is contained in:
Sukriti Sharma 2024-12-07 00:02:14 -07:00 • committed by GitHub
parent c5ede3849f
commit 784a14aa49
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
4 changed files with 165 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

8
convert_hf_to_gguf.py
View file

@ -661,6 +661,9 @@ class Model:
if chkhsh == "1431a23e583c97432bc230bff598d103ddb5a1f89960c8f1d1051aaa944d0b35":
# ref: https://huggingface.co/sapienzanlp/Minerva-7B-base-v1.0
res = "minerva-7b"
if chkhsh == "8b5a93ed704057481f240da0be7e7dca721d7f8f4755263b6807227a2cbeae65":
# ref: https://huggingface.co/sentence-transformers/stsb-roberta-base
res = "roberta-bpe"
if res is None:
logger.warning("\n")
@ -2533,7 +2536,7 @@ class InternLM2Model(Model):
return [(self.map_tensor_name(name), data_torch)]
@Model.register("BertModel", "CamembertModel")
@Model.register("BertModel", "CamembertModel", "RobertaModel")
class BertModel(Model):
model_arch = gguf.MODEL_ARCH.BERT
@ -2574,7 +2577,8 @@ class BertModel(Model):
# we need this to validate the size of the token_type embeddings
# though currently we are passing all zeros to the token_type embeddings
self.gguf_writer.add_token_type_count(2) # "Sequence A" or "Sequence B"
# "Sequence A" or "Sequence B"
self.gguf_writer.add_token_type_count(self.hparams.get("type_vocab_size", 1))
# convert to phantom space vocab
def phantom(tok):