423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-08-23 14:22:51 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix qwen eos token

Browse Source 
Former-commit-id: e21ae0135610bad8116cadbe4b184aac8e279d7c
This commit is contained in:
hiyouga 2023-08-06 13:31:17 +08:00
parent 39955c28ff
commit 2faa1af4eb
1 changed files with 3 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
src/llmtuner/tuner/core/loader.py
View File

@ -68,7 +68,9 @@ def load_model_and_tokenizer(
padding_side=model_args.padding_side,
**config_kwargs
)
if tokenizer.pad_token_id is None or tokenizer.pad_token_id == 64000: # 64000 for baichuan model (older version)
if tokenizer.eos_token_id is None: # fix qwen tokenizer
tokenizer.eos_token = "<|endoftext|>"
if tokenizer.pad_token_id is None: # add pad token
tokenizer.pad_token = tokenizer.eos_token
config = AutoConfig.from_pretrained(model_args.model_name_or_path, **config_kwargs)