Changes LLMRayActor to use vllm's AsyncLLMEngine #1016
Draft
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
…tures are complete.
…list[CompletionOutput].
…ration The issue was that after a tool call, we would loop back and try to generate again with the same sub_request_id. This caused vLLM to reject or hang on the duplicate request ID. Now we append '_iterN' to create unique IDs for each generation attempt within the same request.
finbarrtimbers
commented
Oct 7, 2025
| refs = [ | ||
| engine.update_weight.remote( | ||
| name, dtype=param.dtype, shape=shape, empty_cache=count == num_params | ||
| name, dtype=str(param.dtype), shape=shape, empty_cache=count == num_params |
There was a problem hiding this comment.
This is needed to support the async engine's serialization, which for some unknown reason, goes through MessagePack, and not pickle.
finbarrtimbers
commented
Oct 7, 2025
|
|
||
| assert dtype == self.model_config.dtype, f"mismatch dtype: src {dtype}, dst {self.model_config.dtype}" | ||
| weight = torch.empty(shape, dtype=dtype, device="cuda") | ||
| assert dtype == str(self.model_config.dtype), ( |
There was a problem hiding this comment.
As discussed earlier, this is done because of weird serialization format in vLLM's async engine.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
By moving to the async engine, we have a much cleaner central processing loop, as we process each request in a dedicated function, relying on vllm to handle the batch processing.
SGLang and TensorRT-LLM both use async engines, so by moving to VLLM's async API, it will be much easier to test out those engines.
Runs with
inflight_updates=True:And with
inflight_updates=False: