wip

Signed-off-by: Woosuk Kwon <woosuk@thinkingmachines.ai>

vllm/v1/worker/gpu/model_runner.py

@@ -402,6 +402,11 @@ class GPUModelRunner:
             sampling_metadata=sampling_metadata,
         )
 
+        needs_prompt_logprobs = np.any(
+            self.req_states.needs_prompt_logprobs[idx_mapping_np])
+        if needs_prompt_logprobs:
+            pass
+
         if use_dp_sampler:
             # All-gather the outputs.
             sampler_output = all_gather_sampler_output(