Qwen

/

Qwen3-Reranker-0.6B

like 280

Follow

Qwen 63.3k

This PR proposes improvements to the vLLM usage example:

1. Updates the instruction and query template to match the format used in the Transformers example (removes unnecessary newlines).

2. Fixes a bug in input creation procedure: Sets add_generation_prompt=True in apply_chat_template and removes the suffix and suffix_tokens variables.
   Previously, the combination of <|im_end|>\n tokens was added twice: once by apply_chat_template and again via suffix_tokens, which resulted in inconsistent input strings.