feat(model): Support glm4.5 models (#2867)

examples/agents/auto_plan_agent_dialogue_example.py

@@ -35,8 +35,6 @@ initialize_tracer(
 async def main():
     from dbgpt.model.proxy.llms.siliconflow import SiliconFlowLLMClient
 
-    agent_memory = AgentMemory()
-
     llm_client = SiliconFlowLLMClient(
         model_alias=os.getenv(
             "SILICONFLOW_MODEL_VERSION", "Qwen/Qwen2.5-Coder-32B-Instruct"

packages/dbgpt-core/pyproject.toml

@@ -121,7 +121,7 @@ llama_cpp_server = [
     "llama-cpp-server-py>=0.1.4",
 ]
 proxy_ollama = ["ollama"]
-proxy_zhipuai = ["zhipuai>=2.1.5"]
+proxy_zhipuai = ["openai>=1.59.6"]
 proxy_tongyi = [
     # tongyi supported by openai package
     "openai",

packages/dbgpt-core/src/dbgpt/model/proxy/llms/zhipu.py

@@ -2,9 +2,9 @@ import logging
 import os
 from concurrent.futures import Executor
 from dataclasses import dataclass, field
-from typing import Iterator, Optional, Type, Union
+from typing import TYPE_CHECKING, Any, Dict, Optional, Type, Union
 
-from dbgpt.core import MessageConverter, ModelMetadata, ModelOutput, ModelRequest
+from dbgpt.core import ModelMetadata
 from dbgpt.core.awel.flow import (
     TAGS_ORDER_HIGH,
     ResourceCategory,
@@ -14,25 +14,30 @@ from dbgpt.core.interface.parameter import LLMDeployModelParameters
 from dbgpt.model.proxy.base import (
     AsyncGenerateStreamFunction,
     GenerateStreamFunction,
-    ProxyLLMClient,
     register_proxy_model_adapter,
 )
 from dbgpt.model.proxy.llms.proxy_model import ProxyModel, parse_model_request
 from dbgpt.util.i18n_utils import _
 
-from .chatgpt import OpenAICompatibleDeployModelParameters
+from .chatgpt import OpenAICompatibleDeployModelParameters, OpenAILLMClient
 
-_DEFAULT_MODEL = "glm-4-plus"
+if TYPE_CHECKING:
+    from httpx._types import ProxiesTypes
+    from openai import AsyncAzureOpenAI, AsyncOpenAI
+
+    ClientType = Union[AsyncAzureOpenAI, AsyncOpenAI]
 
 logger = logging.getLogger(__name__)
 
+_DEFAULT_MODEL = "glm-4.5"
+
 
 @auto_register_resource(
     label=_("Zhipu Proxy LLM"),
     category=ResourceCategory.LLM_CLIENT,
     tags={"order": TAGS_ORDER_HIGH},
     description=_("Zhipu proxy LLM configuration."),
-    documentation_url="https://open.bigmodel.cn/dev/api/normal-model/glm-4#overview",
+    documentation_url="https://docs.bigmodel.cn/cn/guide/start/model-overview",
     show_in_ui=False,
 )
 @dataclass
@@ -57,52 +62,32 @@ class ZhipuDeployModelParameters(OpenAICompatibleDeployModelParameters):
     )
 
 
-def zhipu_generate_stream(
+async def zhipu_generate_stream(
     model: ProxyModel, tokenizer, params, device, context_len=2048
 ):
-    """Zhipu ai, see: https://open.bigmodel.cn/dev/api#overview"""
-    model_params = model.get_params()
-    print(f"Model: {model}, model_params: {model_params}")
-
-    # TODO: Support convert_to_compatible_format config, zhipu not support system
-    #  message
-    # convert_to_compatible_format = params.get("convert_to_compatible_format", False)
-    # history, systems = __convert_2_zhipu_messages(messages)
+    """Zhipu ai, see: https://docs.bigmodel.cn/cn/guide/start/model-overview"""
     client: ZhipuLLMClient = model.proxy_llm_client
     request = parse_model_request(params, client.default_model, stream=True)
-    for r in client.sync_generate_stream(request):
+    async for r in client.generate_stream(request):
         yield r
 
 
-class ZhipuLLMClient(ProxyLLMClient):
+class ZhipuLLMClient(OpenAILLMClient):
     def __init__(
         self,
         model: Optional[str] = _DEFAULT_MODEL,
         api_key: Optional[str] = None,
         api_base: Optional[str] = None,
+        api_type: Optional[str] = None,
+        api_version: Optional[str] = None,
+        proxies: Optional["ProxiesTypes"] = None,
         model_alias: Optional[str] = _DEFAULT_MODEL,
+        timeout: Optional[int] = 240,
         context_length: Optional[int] = 8192,
-        executor: Optional[Executor] = None,
+        openai_client: Optional["ClientType"] = None,
+        openai_kwargs: Optional[Dict[str, Any]] = None,
+        **kwargs,
     ):
-        try:
-            from zhipuai import ZhipuAI
-
-        except ImportError as exc:
-            if (
-                "No module named" in str(exc)
-                or "cannot find module" in str(exc).lower()
-            ):
-                raise ValueError(
-                    "The python package 'zhipuai' is not installed. "
-                    "Please install it by running `pip install zhipuai`."
-                ) from exc
-            else:
-                raise ValueError(
-                    "Could not import python package: zhipuai "
-                    "This may be due to a version that is too low. "
-                    "Please upgrade the zhipuai package by running "
-                    "`pip install --upgrade zhipuai`."
-                ) from exc
         if not model:
             model = _DEFAULT_MODEL
         if not api_key:
@@ -111,13 +96,19 @@ class ZhipuLLMClient(ProxyLLMClient):
 
         api_key = self._resolve_env_vars(api_key)
         api_base = self._resolve_env_vars(api_base)
-        self._model = model
-        self.client = ZhipuAI(api_key=api_key, base_url=api_base)
-
         super().__init__(
-            model_names=[model, model_alias],
+            api_key=api_key,
+            api_base=api_base,
+            api_type=api_type,
+            api_version=api_version,
+            model=model,
+            proxies=proxies,
+            timeout=timeout,
+            model_alias=model_alias,
             context_length=context_length,
-            executor=executor,
+            openai_client=openai_client,
+            openai_kwargs=openai_kwargs,
+            **kwargs,
         )
 
     @classmethod
@@ -147,55 +138,20 @@ class ZhipuLLMClient(ProxyLLMClient):
 
     @property
     def default_model(self) -> str:
-        return self._model
-
-    def sync_generate_stream(
-        self,
-        request: ModelRequest,
-        message_converter: Optional[MessageConverter] = None,
-    ) -> Iterator[ModelOutput]:
-        request = self.local_covert_message(request, message_converter)
-
-        messages = request.to_common_messages(support_system_role=False)
-
-        model = request.model or self._model
-        try:
-            logger.debug(
-                f"Send request to zhipu ai, model: {model}, request: {request}"
-            )
-            response = self.client.chat.completions.create(
-                model=model,
-                messages=messages,
-                temperature=request.temperature,
-                max_tokens=request.max_new_tokens,
-                top_p=request.top_p,
-                stream=True,
-            )
-            partial_text = ""
-            for chunk in response:
-                if not chunk.choices or not chunk.choices[0].delta:
-                    continue
-                delta_content = chunk.choices[0].delta.content
-                finish_reason = chunk.choices[0].finish_reason
-                partial_text += delta_content
-                if logger.isEnabledFor(logging.DEBUG):
-                    print(delta_content, end="")
-                yield ModelOutput(
-                    text=partial_text, error_code=0, finish_reason=finish_reason
-                )
-            if not partial_text:
-                yield ModelOutput(text="**LLMServer Generate Empty.**", error_code=1)
-
-        except Exception as e:
-            yield ModelOutput(
-                text=f"**LLMServer Generate Error, Please CheckErrorInfo.**: {e}",
-                error_code=1,
-            )
+        return _DEFAULT_MODEL
 
 
 register_proxy_model_adapter(
     ZhipuLLMClient,
     supported_models=[
+        ModelMetadata(
+            model=["glm-4.5", "glm-4.5-air", "glm-4.5-x", "glm-4.5-airx"],
+            context_length=128 * 1024,
+            max_output_length=96 * 1024,
+            description="GLM-4.5 by Zhipu AI",
+            link="https://docs.bigmodel.cn/cn/guide/start/model-overview",
+            function_calling=True,
+        ),
         ModelMetadata(
             model=["glm-4-plus", "glm-4-air", "glm-4-air-0111"],
             context_length=128 * 1024,