remove deprecated codes (#3759)

* remove deprecated codes

* update

* update

* remote tritonserver backend

Author: lvhan028

lmdeploy/archs.py

@@ -31,35 +31,25 @@ def autoget_backend(model_path: str) -> Literal['turbomind', 'pytorch']:
     Returns:
         str: the backend type.
     """
-    from lmdeploy.pytorch.supported_models import is_supported as is_supported_pytorch
 
-    pytorch_has, turbomind_has = False, False
+    turbomind_has = False
     is_turbomind_installed = True
     try:
         from lmdeploy.turbomind.supported_models import is_supported as is_supported_turbomind
         turbomind_has = is_supported_turbomind(model_path)
     except ImportError:
         is_turbomind_installed = False
 
-    pytorch_has = is_supported_pytorch(model_path)
-
-    try_run_msg = (f'Try to run with pytorch engine because `{model_path}`'
-                   ' is not explicitly supported by lmdeploy. ')
     if is_turbomind_installed:
         if not turbomind_has:
-            if pytorch_has:
-                logger.warning('Fallback to pytorch engine because '
-                               f'`{model_path}` not supported by turbomind'
-                               ' engine.')
-            else:
-                logger.warning(try_run_msg)
+            logger.warning('Fallback to pytorch engine because '
+                           f'`{model_path}` not supported by turbomind'
+                           ' engine.')
     else:
         logger.warning('Fallback to pytorch engine because turbomind engine is not '
                        'installed correctly. If you insist to use turbomind engine, '
                        'you may need to reinstall lmdeploy from pypi or build from '
                        'source and try again.')
-        if not pytorch_has:
-            logger.warning(try_run_msg)
 
     backend = 'turbomind' if turbomind_has else 'pytorch'
     return backend

lmdeploy/pytorch/chat.py

@@ -6,8 +6,7 @@
 from typing import Optional
 
 from lmdeploy.messages import GenerationConfig, PytorchEngineConfig
-from lmdeploy.model import ChatTemplateConfig
-from lmdeploy.serve.async_engine import get_names_from_model
+from lmdeploy.model import ChatTemplateConfig, best_match_model
 
 os.environ['TM_LOG_LEVEL'] = 'ERROR'
 
@@ -85,7 +84,7 @@ async def __chat_step(prompt: str):
     async def __chat_loop(model_path: str):
         """Chat loop."""
         __reset_chat_state()
-        _, chat_template_name = get_names_from_model(model_path)
+        chat_template_name = best_match_model(model_path)
         while True:
             prompt = input_prompt(chat_template_name)
             await __chat_step(prompt)

lmdeploy/pytorch/supported_models.py

@@ -4,7 +4,6 @@
 import atexit
 import concurrent.futures
 import dataclasses
-import os
 import random
 from contextlib import asynccontextmanager, closing
 from copy import deepcopy
@@ -33,23 +32,6 @@
 logger = get_logger('lmdeploy')
 
 
-def get_names_from_model(model_path: str, model_name: str = None):
-    """Get model name and chat template name from workspace model."""
-    triton_model_path = os.path.join(model_path, 'triton_models', 'weights')
-    if not os.path.exists(triton_model_path):
-        chat_template_name = best_match_model(model_path)
-    else:
-        # `model_path` refers to a turbomind model, reading
-        # chat_template_name from the config
-        config_path = os.path.join(triton_model_path, 'config.yaml')
-        with open(config_path, 'r') as f:
-            import yaml
-            config = yaml.safe_load(f)
-        chat_template_name = config['model_config']['chat_template']
-    model_name = model_name if model_name else model_path
-    return model_name, chat_template_name
-
-
 @dataclasses.dataclass
 class GenOut:
     """Pack all response information together."""
@@ -266,7 +248,8 @@ def __init__(self,
         logger.info(f'input backend={backend}, backend_config={backend_config}')
         logger.info(f'input chat_template_config={chat_template_config}')
 
-        self.model_name, chat_template_name = get_names_from_model(model_path, model_name)
+        self.model_name = model_name if model_name else model_path
+        chat_template_name = best_match_model(model_path)
         if chat_template_config is None:
             chat_template_config = ChatTemplateConfig(chat_template_name)
         elif chat_template_config.model_name is None: