[Perf][MRV2] Skip empty draft-token resets on request admission (#61640)

Signed-off-by: Jakub Zakrzewski <jzakrzewski@nvidia.com>
Co-authored-by: OpenAI Codex <noreply@openai.com>
Co-authored-by: Nick Hill <nickhill123@gmail.com>

vllm/v1/worker/gpu/states.py

@@ -114,7 +114,8 @@ class RequestState:
         self.num_computed_tokens_np[req_idx] = num_computed_tokens
         self.num_computed_tokens.stage_write_elem(req_idx, num_computed_tokens)
 
-        self.draft_tokens[req_idx].zero_()
+        if self.num_speculative_steps > 0:
+            self.draft_tokens[req_idx].zero_()
 
     def apply_staged_writes(self) -> None:
         self.prompt_len.copy_to_uva()