优化pp+mtp代码

vllm/model_executor/models/deepseek_mtp.py

@@ -266,6 +266,7 @@ class DeepSeekMTP(nn.Module, DeepseekV2MixtureOfExperts, SupportsPP):
             os.environ['LLAMA_NN'] = '0'
             os.environ['LM_NN'] = '0'
         self.use_llama_nn = os.environ.get('LLAMA_NN') == '1'
+        self.use_pp = vllm_config.parallel_config.pipeline_parallel_size > 1
 
 
     def set_moe_parameters(self):
@@ -351,8 +352,8 @@ class DeepSeekMTP(nn.Module, DeepseekV2MixtureOfExperts, SupportsPP):
 
             spec_layer = get_spec_layer_idx_from_weight_name(self.config, name)
             if spec_layer is None:
-                # load embed_tokens weight from target model if mtp weights missing embed_tokens
-                if "embed_tokens" in name:
+                # load embed_tokens weight from target model in pp mode
+                if "embed_tokens" in name and self.use_pp:
                     for local_name in params_dict.keys():
                         if "embed_tokens" in local_name:
                             param = params_dict[local_name]

vllm/v1/core/sched/scheduler.py

@@ -1657,11 +1657,10 @@ class Scheduler(SchedulerInterface):
         for idx, req in enumerate(itertools.chain(running_reqs, resumed_reqs)):
             req_id = req.request_id
             req_ids.append(req_id)
-            #if self.use_pp:
             # NOTE: In PP+async scheduling, we consume token ids via a direct GPU
             # broadcast path (`input_batch.prev_sampled_token_ids`), so we can
             # omit this payload.
-            if self.use_pp and not self.scheduler_config.async_scheduling:
+            if self.use_pp:
                 # When using PP, the scheduler sends the sampled tokens back,
                 # because there's no direct communication between the first-
                 # stage worker and the last-stage worker. Otherwise, we don't

vllm/v1/worker/gpu_model_runner.py

@@ -931,7 +931,7 @@ class GPUModelRunner(
         The SamplingMetadata is updated and copied to the GPU if there is a
         new/resumed/paused/finished request in the batch.
         """
-        if scheduler_output.total_num_scheduled_tokens == 0:
+        if scheduler_output.total_num_scheduled_tokens == 0 and not self.use_async_scheduling:
             return
         # Remove finished requests from the cached states.
         for req_id in scheduler_output.finished_req_ids:
@@ -1092,13 +1092,9 @@ class GPUModelRunner(
             req_state.num_computed_tokens = num_computed_tokens
 
             if not is_last_rank:
-                if not req_data.new_token_ids:
-                    # Async scheduled PP: Sampled tokens propagated via GPU broadcast.
-                    new_token_ids: list[int] = []
-                else:
-                    # Non-async scheduling with PP: The scheduler sends
-                    # sampled token ids back because there's no direct communication
-                    # between the first-stage worker and the last-stage worker.
+                # When using PP, the scheduler sends the sampled tokens back,
+                # because there's no direct communication between the first-
+                # stage worker and the last-stage worker.
                 new_token_ids = req_data.new_token_ids[i]
                 # Add the sampled token(s) from the previous step (if any).
                 # This doesn't include "unverified" tokens like spec tokens.
@@ -1109,9 +1105,7 @@ class GPUModelRunner(
                     # Avoid slicing list in most common case.
                     req_state.output_token_ids.append(new_token_ids[-1])
                 elif num_new_tokens > 0:
-                        req_state.output_token_ids.extend(
-                            new_token_ids[-num_new_tokens:]
-                        )
+                    req_state.output_token_ids.extend(new_token_ids[-num_new_tokens:])
             elif num_output_tokens < len(req_state.output_token_ids):
                 # Some output tokens were discarded due to a sync-KV-load
                 # failure. Align the cached state.