Qwen3Guard-Gen-0.6B chat template error on vLLM despite tokenizer_config.json existing

#8
by Chinofritz - opened

I'm getting a chat template error when deploying Qwen3Guard-Gen-0.6B with vLLM, even though the model has a chat_template defined in tokenizer_config.json.

Setup:

  • vLLM: vllm/vllm-openai:v0.9.0.1
  • Launch flags: --max-model-len 2048 --trust-remote-code --served-model-name Qwen/Qwen3Guard-Gen-0.6B
  • Using LangChain's ChatOpenAI client

Error:

ValueError: As of transformers v4.44, default chat template is no longer allowed, so you must provide a chat template if the tokenizer does not define one

The tokenizer_config.json clearly has a chat_template defined.
Has anyone successfully run Qwen3Guard-Gen with vLLM? Do these models need special handling for the chat template?

Thanks!

Sign up or log in to comment