update logging

Author: tedzhouhk

examples/vllm_v0/components/worker.py

@@ -202,7 +202,6 @@ async def callback(request: RemotePrefillRequest):
 
     @endpoint()
     async def generate(self, request: PreprocessedRequest):
-        print(request)
         request_id = str(uuid.uuid4())
 
         if self.disaggregated_router is not None:
@@ -213,7 +212,7 @@ async def generate(self, request: PreprocessedRequest):
                 prefill_queue_size = await prefill_queue.get_queue_size()
             disagg_router_decision = await self.disaggregated_router.prefill_remote(
                 len(request.token_ids),
-                0,  # TODO: return prefix hit rate from dynamo-run router
+                request.estimated_prefix_hit_num_blocks * self.engine_args.block_size,
                 prefill_queue_size,
             )
         else:
@@ -226,12 +225,12 @@ async def generate(self, request: PreprocessedRequest):
                 remote_prefill_request_callback=self.get_remote_prefill_request_callback(),
             )
             logger.info(
-                f"Prefilling remotely for request {request_id} with length {len(request.token_ids)}"
+                f"Prefilling remotely for request {request_id} with length {len(request.token_ids)} (estimated prefix hit length {request.estimated_prefix_hit_num_blocks * self.engine_args.block_size})"
             )
         else:
             remote_prefill_params = None
             logger.info(
-                f"Prefilling locally for request {request_id} with length {len(request.token_ids)}"
+                f"Prefilling locally for request {request_id} with length {len(request.token_ids)} (estimated prefix hit length {request.estimated_prefix_hit_num_blocks * self.engine_args.block_size})"
             )
 
         sampling_params = SamplingParams(**self.default_sampling_params)