[Frontend] [Core] feat: Add model loading using tensorizer (#3476)

vllm/engine/llm_engine.py

@@ -6,7 +6,7 @@ from transformers import PreTrainedTokenizer
 import vllm
 from vllm.config import (CacheConfig, DeviceConfig, LoRAConfig, ModelConfig,
                          ParallelConfig, SchedulerConfig, SpeculativeConfig,
-                         VisionLanguageConfig)
+                         TensorizerConfig, VisionLanguageConfig)
 from vllm.core.scheduler import Scheduler, SchedulerOutputs
 from vllm.engine.arg_utils import EngineArgs
 from vllm.engine.metrics import StatLogger, Stats
@@ -74,6 +74,7 @@ class LLMEngine:
         lora_config: Optional[LoRAConfig],
         vision_language_config: Optional[VisionLanguageConfig],
         speculative_config: Optional[SpeculativeConfig],
+        tensorizer_config: Optional[TensorizerConfig],
         executor_class: Type[ExecutorBase],
         log_stats: bool,
         usage_context: UsageContext = UsageContext.ENGINE_CONTEXT,
@@ -110,6 +111,7 @@ class LLMEngine:
         self.scheduler_config = scheduler_config
         self.device_config = device_config
         self.speculative_config = speculative_config
+        self.tensorizer_config = tensorizer_config
         self.log_stats = log_stats
 
         self._init_tokenizer()
@@ -125,6 +127,7 @@ class LLMEngine:
             lora_config=lora_config,
             vision_language_config=vision_language_config,
             speculative_config=speculative_config,
+            tensorizer_config=tensorizer_config,
         )
 
         self._initialize_kv_caches()
@@ -264,6 +267,9 @@ class LLMEngine:
     def _verify_args(self) -> None:
         self.model_config.verify_with_parallel_config(self.parallel_config)
         self.cache_config.verify_with_parallel_config(self.parallel_config)
+        if self.tensorizer_config:
+            self.tensorizer_config.verify_with_parallel_config(
+                self.parallel_config)
         if self.lora_config:
             self.lora_config.verify_with_model_config(self.model_config)
             self.lora_config.verify_with_scheduler_config(