We don't publish
your competitive advantage.

IFCustom chat template or chat_template_kwargs are ignored for GPT-OSS models; hardcoded system message is used instead.

THENModify the `get_system_message()` function in `vllm/entrypoints/openai/parser/harmony_utils.py` to accept and apply `chat_template_kwargs`, or patch the function at runtime to override the hardcoded defaults such as `model_identity`. Alternatively, pass custom values directly in the request payload rather than relying on `chat_template_kwargs`.

IFChat template for gpt-oss models ignores user-provided chat_template_kwargs and model identity, resulting in a hardcoded system prompt even after modifying the template.

THENUpdate the `get_system_message()` call in `vllm/entrypoints/openai/serving_chat.py` to pass `chat_template_kwargs` and any user-specified model identity. Also ensure `get_system_message()` in `vllm/entrypoints/openai/parser/harmony_utils.py` accepts and uses these parameters instead of falling back to hardcoded defaults.

IFChat template customization or chat_template_kwargs (e.g., model_identity, reasoning_effort) are ignored for gpt-oss models in vLLM.

THENModify the get_system_message function in vllm/entrypoints/openai/parser/harmony_utils.py to accept and apply chat_template_kwargs, or directly edit the function's default parameters (e.g., model_identity). Alternatively, avoid using the harmony path by switching to a non-gpt-oss model or applying the template via the standard apply_hf_chat_template flow.