[Model]: Add transformers backend support (#11330)

# Adds support for `transformers` as a backend

Following https://github.com/huggingface/transformers/pull/35235, a
bunch of models should already be supported, we are ramping up support
for more models.

Thanks @Isotr0py for the TP support, and @hmellor for his help as well!
This includes: 
- `trust_remote_code=True` support: any model on the hub, if it
implements attention the correct way can be natively supported!!
- tensor parallel support

---------

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Signed-off-by: Isotr0py <2037008807@qq.com>
Co-authored-by: Isotr0py <41363108+Isotr0py@users.noreply.github.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Isotr0py <2037008807@qq.com>
Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com>
Co-authored-by: Michael Goin <mgoin64@gmail.com>
Co-authored-by: Isotr0py <mozf@mail2.sysu.edu.cn>

vllm/engine/arg_utils.py

@@ -13,10 +13,10 @@ import vllm.envs as envs
 from vllm.config import (CacheConfig, CompilationConfig, ConfigFormat,
                          DecodingConfig, DeviceConfig, HfOverrides,
                          KVTransferConfig, LoadConfig, LoadFormat, LoRAConfig,
-                         ModelConfig, ObservabilityConfig, ParallelConfig,
-                         PoolerConfig, PromptAdapterConfig, SchedulerConfig,
-                         SpeculativeConfig, TaskOption, TokenizerPoolConfig,
-                         VllmConfig)
+                         ModelConfig, ModelImpl, ObservabilityConfig,
+                         ParallelConfig, PoolerConfig, PromptAdapterConfig,
+                         SchedulerConfig, SpeculativeConfig, TaskOption,
+                         TokenizerPoolConfig, VllmConfig)
 from vllm.executor.executor_base import ExecutorBase
 from vllm.logger import init_logger
 from vllm.model_executor.layers.quantization import QUANTIZATION_METHODS
@@ -199,6 +199,7 @@ class EngineArgs:
     generation_config: Optional[str] = None
     override_generation_config: Optional[Dict[str, Any]] = None
     enable_sleep_mode: bool = False
+    model_impl: str = "auto"
 
     calculate_kv_scales: Optional[bool] = None
 
@@ -378,6 +379,18 @@ class EngineArgs:
             'qualified names that can be passed with the `logits_processors` '
             'extra completion argument. Defaults to None, which allows no '
             'processors.')
+        parser.add_argument(
+            '--model-impl',
+            type=str,
+            default=EngineArgs.model_impl,
+            choices=[f.value for f in ModelImpl],
+            help='Which implementation of the model to use.\n\n'
+            '* "auto" will try to use the vLLM implementation if it exists '
+            'and fall back to the Transformers implementation if no vLLM '
+            'implementation is available.\n'
+            '* "vllm" will use the vLLM model implementation.\n'
+            '* "transformers" will use the Transformers model '
+            'implementation.\n')
         # Parallel arguments
         parser.add_argument(
             '--distributed-executor-backend',
@@ -1017,6 +1030,7 @@ class EngineArgs:
             generation_config=self.generation_config,
             override_generation_config=self.override_generation_config,
             enable_sleep_mode=self.enable_sleep_mode,
+            model_impl=self.model_impl,
         )
 
     def create_load_config(self) -> LoadConfig: