Fixing P3L incompatibility with cython. (#200)

* Fixing incompatibility with cython.

* Change type as per reviewer suggestions

Co-authored-by: Gregory Shtrasberg <[email protected]>

---------

Co-authored-by: Matt Wong <[email protected]>
Co-authored-by: Gregory Shtrasberg <[email protected]>

Author: 3 people

vllm/model_executor/layers/sampler.py

@@ -1436,7 +1436,8 @@ def _build_sampler_output(
         deferred_sample_results_args=deferred_sample_results_args)
 
 
-def _get_next_prompt_tokens(seq_group: SequenceGroupToSample) -> List[int]:
+def _get_next_prompt_tokens(
+        seq_group: SequenceGroupToSample) -> Tuple[int, ...]:
     """Get a list of next prompt tokens to compute logprob from a
         given sequence group.