[Feature][Executor] GPU Model Runner Supports prompt_logprobs and max_logprobs #4769
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Thanks for your contribution! |
yuanlehome
reviewed
Nov 4, 2025
Comment on lines
568
to
569
| if request.sampling_params.prompt_logprobs is not None: | ||
| self.prompt_logprobs_reqs[request.request_id] = request |
Collaborator
There was a problem hiding this comment.
另外RL场景中使用的话，需要在clear_requests函数中清空一下model_runner的一些对象，包括这个，也顺便梳理下有没有其他的对象需要清除
Collaborator
Author
There was a problem hiding this comment.
prompt变长和增大并发都会导致显存增长，但不存在显存泄露，这是符合预期的。
gongshaotian
reviewed
Nov 4, 2025
gongshaotian
reviewed
Nov 4, 2025
Comment on lines
+590
to
+591
| self.prompt_logprobs_reqs.pop(request.request_id, None) | ||
| self.in_progress_prompt_logprobs.pop(request.request_id, None) |
Collaborator
There was a problem hiding this comment.
抢占时不需要 del self.prompt_logprobs_reqs[req.request_id] 吗
Collaborator
Author
There was a problem hiding this comment.
上面也有抢占时prompt_logprobs_reqs清除的逻辑啊。
ckl117
changed the title
ckl117
changed the title
gongshaotian
reviewed
Nov 4, 2025
Comment on lines
2756
to
2758
| if isinstance(prompt_token_ids, np.ndarray): | ||
| prompt_token_ids = prompt_token_ids.tolist() | ||
| prompt_token_ids_tensor = paddle.to_tensor(prompt_token_ids, dtype="int64") |
Collaborator
There was a problem hiding this comment.
paddle 不支持 ndarray 直接转tensor吗
yuanlehome
approved these changes
Nov 4, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Motivation
GPUModelRunner supports max_logprobs=-1 and prompt_logprobs.
Modifications
Usage or Command
export FD_USE_GET_SAVE_OUTPUT_V1=1 python -m fastdeploy.entrypoints.openai.api_server \ --model ./ERNIE-4.5-0.3B-PT \ --max-model-len 32768 \ --max-num-seqs 128 \ --tensor-parallel-size 1 \ --enable-logprob \ --max-logprobs -1 \ --no-enable-prefix-caching \Accuracy Tests
TODO: Server layer should support
top_logprobs=-1andprompt_logprobs.Checklist
[FDConfig],[APIServer],[Engine],[Scheduler],[PD Disaggregation],[Executor],[Graph Optimization],[Speculative Decoding],[RL],[Models],[Quantization],[Loader],[OP],[KVCache],[DataProcessor],[BugFix],[Docs],[CI],[Optimization],[Feature],[Benchmark],[Others],[XPU],[HPU],[GCU],[DCU],[Iluvatar],[Metax]]pre-commitbefore commit.releasebranch, make sure the PR has been submitted to thedevelopbranch, then cherry-pick it to thereleasebranch with the[Cherry-Pick]PR tag.