Hope to support Reasoning Outputs like vLLM does.

[zhanghw0354]

Currently, when inferring the DeepSeek R1 model via xLLM, due to the lack of support for the Reasoning Outputs feature similar to that of vLLM (https://docs.vllm.ai/en/stable/features/reasoning_outputs.html), the thinking content and final answer content of the DeepSeek R1 model are mixed together. Users need to extract the thinking content from this mixture. It is hoped that support for Reasoning Outputs can be added like vLLM, allowing users to directly obtain the thinking content from the reasoning_content field.

[linkerzhang]

Yep. A basic function should be added.

btw, @zhanghw0354 feel free to create a PR if you're interested in implementing this for the community please.

[yingxudeng]

Currently, when inferring the DeepSeek R1 model via xLLM, due to the lack of support for the Reasoning Outputs feature similar to that of vLLM (https://docs.vllm.ai/en/stable/features/reasoning_outputs.html), the thinking content and final answer content of the DeepSeek R1 model are mixed together. Users need to extract the thinking content from this mixture. It is hoped that support for Reasoning Outputs can be added like vLLM, allowing users to directly obtain the thinking content from the reasoning_content field.

@zhanghw0354 Interested in creating a PR to add this feature? You're very welcome to do so!