feat(model): support InternLM (#583)

- Close #575
- Fix the bug that the webserver cannot return model instances

pilot/componet.py

@@ -1,13 +1,17 @@
 from __future__ import annotations
 
 from abc import ABC, abstractmethod
-from typing import Type, Dict, TypeVar, Optional, TYPE_CHECKING
+from typing import Type, Dict, TypeVar, Optional, Union, TYPE_CHECKING
+from enum import Enum
+import logging
 import asyncio
 
 # Checking for type hints during runtime
 if TYPE_CHECKING:
     from fastapi import FastAPI
 
+logger = logging.getLogger(__name__)
+
 
 class LifeCycle:
     """This class defines hooks for lifecycle events of a component."""
@@ -37,6 +41,11 @@ async def async_before_stop(self):
         pass
 
 
+class ComponetType(str, Enum):
+    WORKER_MANAGER = "dbgpt_worker_manager"
+    MODEL_CONTROLLER = "dbgpt_model_controller"
+
+
 class BaseComponet(LifeCycle, ABC):
     """Abstract Base Component class. All custom components should extend this."""
 
@@ -80,11 +89,21 @@ def register(self, componet: Type[BaseComponet], *args, **kwargs):
 
     def register_instance(self, instance: T):
         """Register an already initialized component."""
-        self.componets[instance.name] = instance
+        name = instance.name
+        if isinstance(name, ComponetType):
+            name = name.value
+        if name in self.componets:
+            raise RuntimeError(
+                f"Componse name {name} already exists: {self.componets[name]}"
+            )
+        logger.info(f"Register componet with name {name} and instance: {instance}")
+        self.componets[name] = instance
         instance.init_app(self)
 
-    def get_componet(self, name: str, componet_type: Type[T]) -> T:
+    def get_componet(self, name: Union[str, ComponetType], componet_type: Type[T]) -> T:
         """Retrieve a registered component by its name and type."""
+        if isinstance(name, ComponetType):
+            name = name.value
         component = self.componets.get(name)
         if not component:
             raise ValueError(f"No component found with name {name}")

pilot/configs/model_config.py

@@ -69,6 +69,9 @@ def get_device() -> str:
     # (Llama2 based) We only support WizardLM-13B-V1.2 for now, which is trained from Llama-2 13b, see https://huggingface.co/WizardLM/WizardLM-13B-V1.2
     "wizardlm-13b": os.path.join(MODEL_PATH, "WizardLM-13B-V1.2"),
     "llama-cpp": os.path.join(MODEL_PATH, "ggml-model-q4_0.bin"),
+    # https://huggingface.co/internlm/internlm-chat-7b-v1_1, 7b vs 7b-v1.1: https://github.com/InternLM/InternLM/issues/288
+    "internlm-7b": os.path.join(MODEL_PATH, "internlm-chat-7b-v1_1"),
+    "internlm-7b-8k": os.path.join(MODEL_PATH, "internlm-chat-7b-8k"),
 }
 
 EMBEDDING_MODEL_CONFIG = {

pilot/model/adapter.py

@@ -411,6 +411,29 @@ def loader(self, model_path: str, from_pretrained_kwargs: dict):
         return model, tokenizer
 
 
+class InternLMAdapter(BaseLLMAdaper):
+    """The model adapter for internlm/internlm-chat-7b"""
+
+    def match(self, model_path: str):
+        return "internlm" in model_path.lower()
+
+    def loader(self, model_path: str, from_pretrained_kwargs: dict):
+        revision = from_pretrained_kwargs.get("revision", "main")
+        model = AutoModelForCausalLM.from_pretrained(
+            model_path,
+            low_cpu_mem_usage=True,
+            trust_remote_code=True,
+            **from_pretrained_kwargs,
+        )
+        model = model.eval()
+        if "8k" in model_path.lower():
+            model.config.max_sequence_length = 8192
+        tokenizer = AutoTokenizer.from_pretrained(
+            model_path, use_fast=False, trust_remote_code=True, revision=revision
+        )
+        return model, tokenizer
+
+
 register_llm_model_adapters(VicunaLLMAdapater)
 register_llm_model_adapters(ChatGLMAdapater)
 register_llm_model_adapters(GuanacoAdapter)
@@ -421,6 +444,7 @@ def loader(self, model_path: str, from_pretrained_kwargs: dict):
 register_llm_model_adapters(BaichuanAdapter)
 register_llm_model_adapters(WizardLMAdapter)
 register_llm_model_adapters(LlamaCppAdapater)
+register_llm_model_adapters(InternLMAdapter)
 # TODO Default support vicuna, other model need to tests and Evaluate
 
 # just for test_py, remove this later

pilot/model/cluster/controller/controller.py

@@ -4,6 +4,7 @@
 from typing import List
 
 from fastapi import APIRouter, FastAPI
+from pilot.componet import BaseComponet, ComponetType, SystemApp
 from pilot.model.base import ModelInstance
 from pilot.model.parameter import ModelControllerParameters
 from pilot.model.cluster.registry import EmbeddedModelRegistry, ModelRegistry
@@ -14,7 +15,12 @@
 )
 
 
-class BaseModelController(ABC):
+class BaseModelController(BaseComponet, ABC):
+    name = ComponetType.MODEL_CONTROLLER
+
+    def init_app(self, system_app: SystemApp):
+        pass
+
     @abstractmethod
     async def register_instance(self, instance: ModelInstance) -> bool:
         """Register a given model instance"""
@@ -25,7 +31,7 @@ async def deregister_instance(self, instance: ModelInstance) -> bool:
 
     @abstractmethod
     async def get_all_instances(
-        self, model_name: str, healthy_only: bool = False
+        self, model_name: str = None, healthy_only: bool = False
     ) -> List[ModelInstance]:
         """Fetch all instances of a given model. Optionally, fetch only the healthy instances."""
 
@@ -51,7 +57,7 @@ async def deregister_instance(self, instance: ModelInstance) -> bool:
         return await self.registry.deregister_instance(instance)
 
     async def get_all_instances(
-        self, model_name: str, healthy_only: bool = False
+        self, model_name: str = None, healthy_only: bool = False
     ) -> List[ModelInstance]:
         logging.info(
             f"Get all instances with {model_name}, healthy_only: {healthy_only}"
@@ -94,7 +100,7 @@ async def get_all_model_instances(self) -> List[ModelInstance]:
 
     @sync_api_remote(path="/api/controller/models")
     def sync_get_all_instances(
-        self, model_name: str, healthy_only: bool = False
+        self, model_name: str = None, healthy_only: bool = False
     ) -> List[ModelInstance]:
         pass
 
@@ -110,7 +116,7 @@ async def deregister_instance(self, instance: ModelInstance) -> bool:
         return await self.backend.deregister_instance(instance)
 
     async def get_all_instances(
-        self, model_name: str, healthy_only: bool = False
+        self, model_name: str = None, healthy_only: bool = False
     ) -> List[ModelInstance]:
         return await self.backend.get_all_instances(model_name, healthy_only)
 

pilot/model/conversation.py

@@ -2,6 +2,8 @@
 Fork from fastchat: https://github.com/lm-sys/FastChat/blob/main/fastchat/conversation.py
 
 Conversation prompt templates.
+
+TODO Using fastchat core package
 """
 
 import dataclasses
@@ -366,4 +368,21 @@ def get_conv_template(name: str) -> Conversation:
     )
 )
 
+# Internlm-chat template
+register_conv_template(
+    Conversation(
+        name="internlm-chat",
+        system="A chat between a curious <|User|> and an <|Bot|>. The <|Bot|> gives helpful, detailed, and polite answers to the <|User|>'s questions.\n\n",
+        roles=("<|User|>", "<|Bot|>"),
+        messages=(),
+        offset=0,
+        sep_style=SeparatorStyle.CHATINTERN,
+        sep="<eoh>",
+        sep2="<eoa>",
+        stop_token_ids=[1, 103028],
+        stop_str="<eoa>",
+    )
+)
+
+
 # TODO Support other model conversation template

pilot/openapi/api_v1/api_v1.py

@@ -18,6 +18,7 @@
 from typing import List
 import tempfile
 
+from pilot.componet import ComponetType
 from pilot.openapi.api_view_model import (
     Result,
     ConversationVo,
@@ -352,20 +353,17 @@ async def chat_completions(dialogue: ConversationVo = Body()):
 async def model_types(request: Request):
     print(f"/controller/model/types")
     try:
-        import httpx
-
-        async with httpx.AsyncClient() as client:
-            base_url = request.base_url
-            response = await client.get(
-                f"{base_url}api/controller/models?healthy_only=true",
-            )
         types = set()
-        if response.status_code == 200:
-            models = json.loads(response.text)
-            for model in models:
-                worker_type = model["model_name"].split("@")[1]
-                if worker_type == "llm":
-                    types.add(model["model_name"].split("@")[0])
+        from pilot.model.cluster.controller.controller import BaseModelController
+
+        controller = CFG.SYSTEM_APP.get_componet(
+            ComponetType.MODEL_CONTROLLER, BaseModelController
+        )
+        models = await controller.get_all_instances(healthy_only=True)
+        for model in models:
+            worker_name, worker_type = model.model_name.split("@")
+            if worker_type == "llm":
+                types.add(worker_name)
         return Result.succ(list(types))
 
     except Exception as e:

pilot/server/chat_adapter.py

@@ -247,6 +247,16 @@ def get_generate_stream_func(self, model_path: str):
         return generate_stream
 
 
+class InternLMChatAdapter(BaseChatAdpter):
+    """The model adapter for internlm/internlm-chat-7b"""
+
+    def match(self, model_path: str):
+        return "internlm" in model_path.lower()
+
+    def get_conv_template(self, model_path: str) -> Conversation:
+        return get_conv_template("internlm-chat")
+
+
 register_llm_model_chat_adapter(VicunaChatAdapter)
 register_llm_model_chat_adapter(ChatGLMChatAdapter)
 register_llm_model_chat_adapter(GuanacoChatAdapter)
@@ -257,6 +267,7 @@ def get_generate_stream_func(self, model_path: str):
 register_llm_model_chat_adapter(BaichuanChatAdapter)
 register_llm_model_chat_adapter(WizardLMChatAdapter)
 register_llm_model_chat_adapter(LlamaCppChatAdapter)
+register_llm_model_chat_adapter(InternLMChatAdapter)
 
 # Proxy model for test and develop, it's cheap for us now.
 register_llm_model_chat_adapter(ProxyllmChatAdapter)

pilot/server/componet_configs.py

@@ -9,10 +9,12 @@
 
 def initialize_componets(system_app: SystemApp, embedding_model_name: str):
     from pilot.model.cluster import worker_manager
+    from pilot.model.cluster.controller.controller import controller
 
     system_app.register(
         RemoteEmbeddingFactory, worker_manager, model_name=embedding_model_name
     )
+    system_app.register_instance(controller)
 
 
 class RemoteEmbeddingFactory(EmbeddingFactory):