support --no-enable-chunked-prefill of v1

vllm/config/model.py

@@ -276,6 +276,9 @@ class ModelConfig:
     override_pooler_config: Optional[Union[dict, PoolerConfig]] = None
     """[DEPRECATED] Use `pooler_config` instead. This field will be removed in
     v0.12.0 or v1.0.0, whichever is sooner."""
+    enable_chunked_prefill: Optional[bool] = None
+    """If True, prefill requests can be chunked based
+    on the remaining max_num_batched_tokens."""
 
     # Multimodal config and init vars
     multimodal_config: Optional[MultiModalConfig] = None
@@ -320,6 +323,7 @@ class ModelConfig:
         factors.append(self.rope_scaling)
         factors.append(self.rope_theta)
         factors.append(self.video_pruning_rate)
+        factors.append(self.enable_chunked_prefill)
 
         # hf_config can control how the model looks!
         try:

vllm/engine/arg_utils.py

@@ -1055,6 +1055,7 @@ class EngineArgs:
             logits_processors=self.logits_processors,
             video_pruning_rate=self.video_pruning_rate,
             io_processor_plugin=self.io_processor_plugin,
+            enable_chunked_prefill=self.enable_chunked_prefill,
         )
 
     def validate_tensorizer_args(self):
@@ -1561,6 +1562,10 @@ class EngineArgs:
         if model_config.runner_type != "pooling":
             self.enable_chunked_prefill = True
             
+            if model_config.enable_chunked_prefill is not None and \
+                model_config.enable_chunked_prefill is False:
+                self.enable_chunked_prefill = False
+
             # TODO: When prefix caching supports prompt embeds inputs, this
             # check can be removed.
             if (self.enable_prompt_embeds
@@ -1584,6 +1589,10 @@ class EngineArgs:
             action = "Enabling" if \
                 incremental_prefill_supported else "Disabling"
                 
+            if model_config.enable_chunked_prefill is not None and \
+                model_config.enable_chunked_prefill is False:
+                self.enable_chunked_prefill = False
+
             if self.enable_chunked_prefill is None:
                 self.enable_chunked_prefill = incremental_prefill_supported
                 logger.info("(%s) chunked prefill by default", action)