phoenix/litellm
1
0
Fork 0
forked from phoenix/litellm-mirror
Code Pull requests Activity

feat - try using hf tokenizer

Browse source
This commit is contained in:
Ishaan Jaff 2024-05-16 10:59:29 -07:00
parent c646b809a6
commit 22ba5fa186
3 changed files with 34 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

13
litellm/utils.py
View file

@ -3860,7 +3860,12 @@ def _select_tokenizer(model: str):
return {"type": "huggingface_tokenizer", "tokenizer": tokenizer}
# default - tiktoken
else:
return {"type": "openai_tokenizer", "tokenizer": encoding}
tokenizer = None
try:
tokenizer = Tokenizer.from_pretrained(model)
return {"type": "huggingface_tokenizer", "tokenizer": tokenizer}
except:
return {"type": "openai_tokenizer", "tokenizer": encoding}
def encode(model="", text="", custom_tokenizer: Optional[dict] = None):
@ -4097,6 +4102,7 @@ def token_counter(
text: Optional[Union[str, List[str]]] = None,
messages: Optional[List] = None,
count_response_tokens: Optional[bool] = False,
return_tokenizer_used: Optional[bool] = False,
):
"""
Count the number of tokens in a given text using a specified model.
@ -4189,7 +4195,10 @@ def token_counter(
)
else:
num_tokens = len(encoding.encode(text, disallowed_special=())) # type: ignore
_tokenizer_type = tokenizer_json["type"]
if return_tokenizer_used:
# used by litellm proxy server -> POST /utils/token_counter
return num_tokens, _tokenizer_type
return num_tokens