[Feature] Can base model start the service via openai for ppl evaluation?

[GenerallyCovetous]

Describe the feature

base model over 7/8b requires multiple gpu parallelism on my gpu, already tested instruct model e.g. qwen2_5-72b-instruct can be used for multi-DP evaluation with multiple gpu via openai method (--max-num-workers 8, the reason for using this method is that --hf-gpu-nums doesn't actually work). To summarize, how can base model use openai service method for ppl evaluation?

Will you implement it?

• I would like to implement this feature and create a PR!