[V1] Aggregate chunked prompt logprobs in model runner (#14875)

Signed-off-by: Nick Hill <nhill@redhat.com>
2025-03-24 09:27:57 -07:00
parent 9cc645141d
commit 3aee6573dc
7 changed files with 68 additions and 44 deletions
--- a/vllm/v1/engine/logprobs.py
+++ b/vllm/v1/engine/logprobs.py
@@ -115,7 +115,6 @@ class LogprobsProcessor:
        num_prompt_tokens, num_logprobs = logprobs.shape

        # Pythonize the torch tensors.
-        # TODO(rob): experiment with doing this in EngineCore?
        prompt_token_ranks = ranks.tolist()
        prompt_logprobs = logprobs.tolist()
        token_ids = token_ids.tolist()