You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
fix: load EOS token from model config.json as fallback
Qwen and other models store eos_token_id in config.json rather than
tokenizer_config.json. Now we check both files, preferring tokenizer_config
but falling back to model config.
This fixes the issue where the model wouldn't stop generating at the
correct EOS token.
0 commit comments