Skip to content

convert_hf : fix Gemma v1 conversion #8597

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Jul 21, 2024
Merged

convert_hf : fix Gemma v1 conversion #8597

Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
5a9cb57
convert_hf : fix Gemma v1 conversion
compilade Jul 19, 2024
50d1a03
convert_hf : fix Gemma v1 not setting BOS and EOS tokens
compilade Jul 20, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
23 changes: 17 additions & 6 deletions convert_hf_to_gguf.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -750,7 +750,8 @@ def _create_vocab_sentencepiece(self):
token_id = int(token_id)
token: str = token_data["content"]
if toktypes[token_id] != SentencePieceTokenTypes.UNUSED:
assert tokens[token_id] == token.encode("utf-8")
if tokens[token_id] != token.encode("utf-8"):
logger.warning(f'replacing token {token_id}: {tokens[token_id].decode("utf-8")!r} -> {token!r}')
if token_data.get("special") or self.does_token_look_special(token):
toktypes[token_id] = SentencePieceTokenTypes.CONTROL
else:
Expand Down Expand Up @@ -2011,7 +2012,8 @@ def set_vocab(self):
token_id = int(token_id)
token = foken_data["content"].encode("utf-8")
if toktypes[token_id] != SentencePieceTokenTypes.UNUSED:
assert tokens[token_id] == token
if tokens[token_id] != token:
logger.warning(f'replacing token {token_id}: {tokens[token_id].decode("utf-8")!r} -> {token.decode("utf-8")!r}')
tokens[token_id] = token
scores[token_id] = -1000.0
toktypes[token_id] = SentencePieceTokenTypes.USER_DEFINED
Expand All @@ -2027,7 +2029,8 @@ def set_vocab(self):
token_id = int(foken_data["id"])
token = foken_data["content"].encode("utf-8")
if toktypes[token_id] != SentencePieceTokenTypes.UNUSED:
assert tokens[token_id] == token
if tokens[token_id] != token:
logger.warning(f'replacing token {token_id}: {tokens[token_id].decode("utf-8")!r} -> {token.decode("utf-8")!r}')
tokens[token_id] = token
scores[token_id] = -1000.0
toktypes[token_id] = SentencePieceTokenTypes.USER_DEFINED
Expand Down Expand Up @@ -2266,7 +2269,8 @@ def set_vocab(self):
chat_eos_token_id = token_id
token = token.encode("utf-8")
if toktypes[token_id] != SentencePieceTokenTypes.UNUSED:
assert(tokens[token_id] == token)
if tokens[token_id] != token:
logger.warning(f'replacing token {token_id}: {tokens[token_id].decode("utf-8")!r} -> {token.decode("utf-8")!r}')
tokens[token_id] = token
scores[token_id] = -1000.0
toktypes[token_id] = SentencePieceTokenTypes.USER_DEFINED
Expand All @@ -2285,7 +2289,8 @@ def set_vocab(self):
chat_eos_token_id = token_id
token = token.encode("utf-8")
if toktypes[token_id] != SentencePieceTokenTypes.UNUSED:
assert(tokens[token_id] == token)
if tokens[token_id] != token:
logger.warning(f'replacing token {token_id}: {tokens[token_id].decode("utf-8")!r} -> {token.decode("utf-8")!r}')
tokens[token_id] = token
scores[token_id] = -1000.0
toktypes[token_id] = SentencePieceTokenTypes.USER_DEFINED
Expand Down Expand Up @@ -2461,7 +2466,13 @@ class GemmaModel(Model):
model_arch = gguf.MODEL_ARCH.GEMMA

def set_vocab(self):
self._set_vocab_sentencepiece()
tokens, scores, toktypes = self._create_vocab_sentencepiece()

self.gguf_writer.add_tokenizer_model("llama")
self.gguf_writer.add_tokenizer_pre("default")
self.gguf_writer.add_token_list(tokens)
self.gguf_writer.add_token_scores(scores)
self.gguf_writer.add_token_types(toktypes)

# TODO: these special tokens should be exported only for the CodeGemma family
special_vocab = gguf.SpecialVocab(self.dir_model, load_merges=False,
Expand Down