[Bug]: Bugs in the new logprobs_mode

[YunruiZhang]

🐛 Describe the bug

The new logprobs_mode has many bugs. When set to logprobs_mode = 'processed_logprobs', it returns the same values as logprobs_mode = 'raw_logprobs'. Additionally, when set to 'processed_logits' or 'raw_logits', it does not produce any meaningful output the logprobs are all -inf except for rank = 1, which gives 0. I am using MODEL_ID = "Qwen/Qwen3-30B-A3B-Thinking-2507-FP8"

[Liu-congo]

if no one has been working on it, i'd like to try to fix it

[Liu-congo]

@YunruiZhang emm could you provide a detail reproduction script for the issue, i've test the logprobs_mode on qwen3-0.6B, it worked fine

[YunruiZhang]

Hi, I think my problem is that the vllm engine falls back to v0 when i am trying to use Qwen/Qwen3-30B-A3B-Thinking-2507-FP8. However when I force to use v1 engine it will give error "NotImplementedError: VLLM_USE_V1=1 is not supported with --kv-cache-dtype." it seems the fp8 kv cache is still not supported for Qwen3.

[Liu-congo]

sounds reasonable. but in issue #15191, it seems to have supported fp8 kvcache, i think there might be some further reason we haven't noticed

[github-actions]

This issue has been automatically marked as stale because it has not had any activity within 90 days. It will be automatically closed if no further activity occurs within 30 days. Leave a comment if you feel this issue should remain open. Thank you!

[github-actions]

This issue has been automatically closed due to inactivity. Please feel free to reopen if you feel it is still relevant. Thank you!