[Fix] Fix typo in resolve_hf_chat_template

[fxmarty-amd]

This makes lm_eval --model vllm fail.

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can either: Add ready label to the PR or enable auto-merge.

🚀

[fxmarty-amd]

#18259 (comment) also results in [rank0]: TypeError: resolve_hf_chat_template() missing 1 required keyword-only argument: 'model_config' in lm-eval-harness due to the new keyword arg.

https://github.com/EleutherAI/lm-evaluation-harness/blob/0126f6d15e6d5f9b93f244b8ece5a44bdbac9c2c/lm_eval/models/vllm_causallms.py#L143

[mgoin]

Thanks for identifying, this bug was introduced in #18098

[fxmarty-amd]

@mgoin Thanks! I'll try to submit a fix in lm-eval-harness for the missing model_config.

I also noticed that sglang in lm-eval-harness does not use this resolve_hf_chat_template. Not sure if prompts are different when evaluation with vllm vs sglang then.

Compare:
https://github.com/EleutherAI/lm-evaluation-harness/blob/0126f6d15e6d5f9b93f244b8ece5a44bdbac9c2c/lm_eval/models/vllm_causallms.py#L214
and
https://github.com/EleutherAI/lm-evaluation-harness/blob/0126f6d15e6d5f9b93f244b8ece5a44bdbac9c2c/lm_eval/models/sglang_causallms.py#L408-L413

[mgoin]

Yes @fxmarty-amd this is intentional for the vllm backend.

@anmarques introduced the resolve_hf_chat_template into vllm to allow us to do text evals on models that contain the chat template within chat_template.json instead of tokenizer_config.json. This is basically copied and pasted the way we handle it in vllm