Diuse/zhenxun_bot
1
0
Fork 0
mirror of https://github.com/zhenxun-org/zhenxun_bot.git synced 2026-10-17 10:10:00 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

✨ feat(llm): 全面重构LLM服务模块，增强多模态与工具支持 (#1953)

Browse Source 
* ✨ feat(llm): 全面重构LLM服务模块，增强多模态与工具支持

🚀 核心功能增强
- 多模型链式调用：新增 `pipeline_chat` 支持复杂任务流处理
- 扩展提供商支持：新增 ARK(火山方舟)、SiliconFlow(硅基流动) 适配器
- 多模态处理增强：支持URL媒体文件下载转换，提升输入灵活性
- 历史对话支持：AI.analyze 方法支持历史消息上下文和可选 UniMessage 参数
- 文本嵌入功能：新增 `embed`、`analyze_multimodal`、`search_multimodal` 等API
- 模型能力系统：新增 `ModelCapabilities` 统一管理模型特性（多模态、工具调用等）

🔧 架构重构与优化
- MCP工具系统重构：配置独立化至 `data/llm/mcp_tools.json`，预置常用工具
- API调用逻辑统一：提取通用 `_perform_api_call` 方法，消除代码重复
- 跨平台兼容：Windows平台MCP工具npx命令自动包装处理
- HTTP客户端增强：兼容不同版本httpx代理配置（0.28+版本适配）

🛠️ API与配置完善
- 统一返回类型：`AI.analyze` 统一返回 `LLMResponse` 类型
- 消息转换工具：新增 `message_to_unimessage` 转换函数
- Gemini适配器增强：URL图片下载编码、动态安全阈值配置
- 缓存管理：新增模型实例缓存和管理功能
- 配置预设：扩展 CommonOverrides 预设配置选项
- 历史管理优化：支持多模态内容占位符替换，提升效率

📚 文档与开发体验
- README全面重写：新增完整使用指南、API参考和架构概览
- 文档内容扩充：补充嵌入模型、缓存管理、工具注册等功能说明
- 日志记录增强：支持详细调试信息输出
- API简化：移除冗余函数，优化接口设计

* 🎨  feat(llm): 统一LLM服务函数文档格式

* ✨ feat(llm): 添加新模型并简化提供者配置加载

* 🚨 auto fix by pre-commit hooks

---------

Co-authored-by: webjoin111 <455457521@qq.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
This commit is contained in:
Rumio
2025-07-08 11:15:15 +08:00
committed by GitHub
co-authored by webjoin111 pre-commit-ci[bot]
parent 1e7ae38684
commit 48cbb2bf1d
24 changed files with 2128 additions and 1338 deletions
Download Patch File Download Diff File Expand all files Collapse all files
zhenxun/services/llm/adapters/__init__.py
-2
View File
@@ -8,7 +8,6 @@ from .base import BaseAdapter, OpenAICompatAdapter, RequestData, ResponseData
from .factory import LLMAdapterFactory, get_adapter_for_api_type, register_adapter
from .gemini import GeminiAdapter
from .openai import OpenAIAdapter
from .zhipu import ZhipuAdapter
LLMAdapterFactory.initialize()
@@ -20,7 +19,6 @@ __all__ = [
"OpenAICompatAdapter",
"RequestData",
"ResponseData",
"ZhipuAdapter",
"get_adapter_for_api_type",
"register_adapter",
]
zhenxun/services/llm/adapters/base.py
+57 -15
View File
@@ -17,6 +17,7 @@ if TYPE_CHECKING:
from ..service import LLMModel
from ..types.content import LLMMessage
from ..types.enums import EmbeddingTaskType
from ..types.models import LLMTool
class RequestData(BaseModel):
@@ -60,7 +61,7 @@ class BaseAdapter(ABC):
"""支持的API类型列表"""
pass
def prepare_simple_request(
async def prepare_simple_request(
self,
model: "LLMModel",
api_key: str,
@@ -86,7 +87,7 @@ class BaseAdapter(ABC):
config = model._generation_config
return self.prepare_advanced_request(
return await self.prepare_advanced_request(
model=model,
api_key=api_key,
messages=messages,
@@ -96,13 +97,13 @@ class BaseAdapter(ABC):
)
@abstractmethod
def prepare_advanced_request(
async def prepare_advanced_request(
self,
model: "LLMModel",
api_key: str,
messages: list["LLMMessage"],
config: "LLMGenerationConfig | None" = None,
tools: list[dict[str, Any]] | None = None,
tools: list["LLMTool"] | None = None,
tool_choice: str | dict[str, Any] | None = None,
) -> RequestData:
"""准备高级请求"""
@@ -238,6 +239,9 @@ class BaseAdapter(ABC):
message = choice.get("message", {})
content = message.get("content", "")
if content:
content = content.strip()
parsed_tool_calls: list[LLMToolCall] | None = None
if message_tool_calls := message.get("tool_calls"):
from ..types.models import LLMToolFunction
@@ -375,7 +379,7 @@ class BaseAdapter(ABC):
if model.temperature is not None:
base_config["temperature"] = model.temperature
if model.max_tokens is not None:
if model.api_type in ["gemini", "gemini_native"]:
if model.api_type == "gemini":
base_config["maxOutputTokens"] = model.max_tokens
else:
base_config["max_tokens"] = model.max_tokens
@@ -401,26 +405,51 @@ class OpenAICompatAdapter(BaseAdapter):
"""
@abstractmethod
def get_chat_endpoint(self) -> str:
def get_chat_endpoint(self, model: "LLMModel") -> str:
"""子类必须实现，返回 chat completions 的端点"""
pass
@abstractmethod
def get_embedding_endpoint(self) -> str:
def get_embedding_endpoint(self, model: "LLMModel") -> str:
"""子类必须实现，返回 embeddings 的端点"""
pass
def prepare_advanced_request(
async def prepare_simple_request(
self,
model: "LLMModel",
api_key: str,
prompt: str,
history: list[dict[str, str]] | None = None,
) -> RequestData:
"""准备简单文本生成请求 - OpenAI兼容API的通用实现"""
url = self.get_api_url(model, self.get_chat_endpoint(model))
headers = self.get_base_headers(api_key)
messages = []
if history:
messages.extend(history)
messages.append({"role": "user", "content": prompt})
body = {
"model": model.model_name,
"messages": messages,
}
body = self.apply_config_override(model, body)
return RequestData(url=url, headers=headers, body=body)
async def prepare_advanced_request(
self,
model: "LLMModel",
api_key: str,
messages: list["LLMMessage"],
config: "LLMGenerationConfig | None" = None,
tools: list[dict[str, Any]] | None = None,
tools: list["LLMTool"] | None = None,
tool_choice: str | dict[str, Any] | None = None,
) -> RequestData:
"""准备高级请求 - OpenAI兼容格式"""
url = self.get_api_url(model, self.get_chat_endpoint())
url = self.get_api_url(model, self.get_chat_endpoint(model))
headers = self.get_base_headers(api_key)
openai_messages = self.convert_messages_to_openai_format(messages)
@@ -430,7 +459,21 @@ class OpenAICompatAdapter(BaseAdapter):
}
if tools:
body["tools"] = tools
openai_tools = []
for tool in tools:
if tool.type == "function" and tool.function:
openai_tools.append({"type": "function", "function": tool.function})
elif tool.type == "mcp" and tool.mcp_session:
if callable(tool.mcp_session):
raise ValueError(
"适配器接收到未激活的 MCP 会话工厂。"
"会话工厂应该在 LLMModel.generate_response 中被激活。"
)
openai_tools.append(
tool.mcp_session.to_api_tool(api_type=self.api_type)
)
if openai_tools:
body["tools"] = openai_tools
if tool_choice:
body["tool_choice"] = tool_choice
@@ -444,7 +487,7 @@ class OpenAICompatAdapter(BaseAdapter):
is_advanced: bool = False,
) -> ResponseData:
"""解析响应 - 直接使用基类的 OpenAI 格式解析"""
_ = model, is_advanced # 未使用的参数
_ = model, is_advanced
return self.parse_openai_response(response_json)
def prepare_embedding_request(
@@ -456,8 +499,8 @@ class OpenAICompatAdapter(BaseAdapter):
**kwargs: Any,
) -> RequestData:
"""准备嵌入请求 - OpenAI兼容格式"""
_ = task_type # 未使用的参数
url = self.get_api_url(model, self.get_embedding_endpoint())
_ = task_type
url = self.get_api_url(model, self.get_embedding_endpoint(model))
headers = self.get_base_headers(api_key)
body = {
@@ -465,7 +508,6 @@ class OpenAICompatAdapter(BaseAdapter):
"input": texts,
}
# 应用额外的配置参数
if kwargs:
body.update(kwargs)
zhenxun/services/llm/adapters/factory.py
-2
View File
@@ -22,10 +22,8 @@ class LLMAdapterFactory:
from .gemini import GeminiAdapter
from .openai import OpenAIAdapter
from .zhipu import ZhipuAdapter
cls.register_adapter(OpenAIAdapter())
cls.register_adapter(ZhipuAdapter())
cls.register_adapter(GeminiAdapter())
@classmethod
zhenxun/services/llm/adapters/gemini.py
+31 -69
View File
@@ -14,7 +14,7 @@ if TYPE_CHECKING:
from ..service import LLMModel
from ..types.content import LLMMessage
from ..types.enums import EmbeddingTaskType
from ..types.models import LLMToolCall
from ..types.models import LLMTool, LLMToolCall
class GeminiAdapter(BaseAdapter):
@@ -38,30 +38,16 @@ class GeminiAdapter(BaseAdapter):
return headers
def prepare_advanced_request(
async def prepare_advanced_request(
self,
model: "LLMModel",
api_key: str,
messages: list["LLMMessage"],
config: "LLMGenerationConfig | None" = None,
tools: list[dict[str, Any]] | None = None,
tools: list["LLMTool"] | None = None,
tool_choice: str | dict[str, Any] | None = None,
) -> RequestData:
"""准备高级请求"""
return self._prepare_request(
model, api_key, messages, config, tools, tool_choice
)
def _prepare_request(
self,
model: "LLMModel",
api_key: str,
messages: list["LLMMessage"],
config: "LLMGenerationConfig | None" = None,
tools: list[dict[str, Any]] | None = None,
tool_choice: str | dict[str, Any] | None = None,
) -> RequestData:
"""准备 Gemini API 请求 - 支持所有高级功能"""
effective_config = config if config is not None else model._generation_config
endpoint = self._get_gemini_endpoint(model, effective_config)
@@ -78,7 +64,8 @@ class GeminiAdapter(BaseAdapter):
system_instruction_parts = [{"text": msg.content}]
elif isinstance(msg.content, list):
system_instruction_parts = [
part.convert_for_api("gemini") for part in msg.content
await part.convert_for_api_async("gemini")
for part in msg.content
]
continue
@@ -87,7 +74,9 @@ class GeminiAdapter(BaseAdapter):
current_parts.append({"text": msg.content})
elif isinstance(msg.content, list):
for part_obj in msg.content:
current_parts.append(part_obj.convert_for_api("gemini"))
current_parts.append(
await part_obj.convert_for_api_async("gemini")
)
gemini_contents.append({"role": "user", "parts": current_parts})
elif msg.role == "assistant" or msg.role == "model":
@@ -95,7 +84,9 @@ class GeminiAdapter(BaseAdapter):
current_parts.append({"text": msg.content})
elif isinstance(msg.content, list):
for part_obj in msg.content:
current_parts.append(part_obj.convert_for_api("gemini"))
current_parts.append(
await part_obj.convert_for_api_async("gemini")
)
if msg.tool_calls:
import json
@@ -154,16 +145,22 @@ class GeminiAdapter(BaseAdapter):
all_tools_for_request = []
if tools:
for tool_item in tools:
if isinstance(tool_item, dict):
if "name" in tool_item and "description" in tool_item:
all_tools_for_request.append(
{"functionDeclarations": [tool_item]}
for tool in tools:
if tool.type == "function" and tool.function:
all_tools_for_request.append(
{"functionDeclarations": [tool.function]}
)
elif tool.type == "mcp" and tool.mcp_session:
if callable(tool.mcp_session):
raise ValueError(
"适配器接收到未激活的 MCP 会话工厂。"
"会话工厂应该在 LLMModel.generate_response 中被激活。"
)
else:
all_tools_for_request.append(tool_item)
else:
all_tools_for_request.append(tool_item)
all_tools_for_request.append(
tool.mcp_session.to_api_tool(api_type=self.api_type)
)
elif tool.type == "google_search":
all_tools_for_request.append({"googleSearch": {}})
if effective_config:
if getattr(effective_config, "enable_grounding", False):
@@ -183,11 +180,7 @@ class GeminiAdapter(BaseAdapter):
logger.debug("隐式启用代码执行工具。")
if all_tools_for_request:
gemini_api_tools = self._convert_tools_to_gemini_format(
all_tools_for_request
)
if gemini_api_tools:
body["tools"] = gemini_api_tools
body["tools"] = all_tools_for_request
final_tool_choice = tool_choice
if final_tool_choice is None and effective_config:
@@ -241,38 +234,6 @@ class GeminiAdapter(BaseAdapter):
return f"/v1beta/models/{model.model_name}:generateContent"
def _convert_tools_to_gemini_format(
self, tools: list[dict[str, Any]]
) -> list[dict[str, Any]]:
"""转换工具格式为Gemini格式"""
gemini_tools = []
for tool in tools:
if tool.get("type") == "function":
func = tool["function"]
gemini_tool = {
"functionDeclarations": [
{
"name": func["name"],
"description": func.get("description", ""),
"parameters": func.get("parameters", {}),
}
]
}
gemini_tools.append(gemini_tool)
elif tool.get("type") == "code_execution":
gemini_tools.append(
{"codeExecution": {"language": tool.get("language", "python")}}
)
elif tool.get("type") == "google_search":
gemini_tools.append({"googleSearch": {}})
elif "googleSearch" in tool:
gemini_tools.append({"googleSearch": tool["googleSearch"]})
elif "codeExecution" in tool:
gemini_tools.append({"codeExecution": tool["codeExecution"]})
return gemini_tools
def _convert_tool_choice_to_gemini(
self, tool_choice_value: str | dict[str, Any]
) -> dict[str, Any]:
@@ -395,10 +356,11 @@ class GeminiAdapter(BaseAdapter):
for category, threshold in custom_safety_settings.items():
safety_settings.append({"category": category, "threshold": threshold})
else:
from ..config.providers import get_gemini_safety_threshold
threshold = get_gemini_safety_threshold()
for category in safety_categories:
safety_settings.append(
{"category": category, "threshold": "BLOCK_MEDIUM_AND_ABOVE"}
)
safety_settings.append({"category": category, "threshold": threshold})
return safety_settings if safety_settings else None
zhenxun/services/llm/adapters/openai.py
+12 -31
View File
@@ -1,12 +1,12 @@
"""
OpenAI API 适配器
支持 OpenAI、DeepSeek 和其他 OpenAI 兼容的 API 服务。
支持 OpenAI、DeepSeek、智谱AI 和其他 OpenAI 兼容的 API 服务。
"""
from typing import TYPE_CHECKING
from .base import OpenAICompatAdapter, RequestData
from .base import OpenAICompatAdapter
if TYPE_CHECKING:
from ..service import LLMModel
@@ -21,37 +21,18 @@ class OpenAIAdapter(OpenAICompatAdapter):
@property
def supported_api_types(self) -> list[str]:
return ["openai", "deepseek", "general_openai_compat"]
return ["openai", "deepseek", "zhipu", "general_openai_compat", "ark"]
def get_chat_endpoint(self) -> str:
def get_chat_endpoint(self, model: "LLMModel") -> str:
"""返回聊天完成端点"""
if model.api_type == "ark":
return "/api/v3/chat/completions"
if model.api_type == "zhipu":
return "/api/paas/v4/chat/completions"
return "/v1/chat/completions"
def get_embedding_endpoint(self) -> str:
"""返回嵌入端点"""
def get_embedding_endpoint(self, model: "LLMModel") -> str:
"""根据API类型返回嵌入端点"""
if model.api_type == "zhipu":
return "/v4/embeddings"
return "/v1/embeddings"
def prepare_simple_request(
self,
model: "LLMModel",
api_key: str,
prompt: str,
history: list[dict[str, str]] | None = None,
) -> RequestData:
"""准备简单文本生成请求 - OpenAI优化实现"""
url = self.get_api_url(model, self.get_chat_endpoint())
headers = self.get_base_headers(api_key)
messages = []
if history:
messages.extend(history)
messages.append({"role": "user", "content": prompt})
body = {
"model": model.model_name,
"messages": messages,
}
body = self.apply_config_override(model, body)
return RequestData(url=url, headers=headers, body=body)
zhenxun/services/llm/adapters/zhipu.py
-57
View File
@@ -1,57 +0,0 @@
"""
智谱 AI API 适配器
支持智谱 AI 的 GLM 系列模型，使用 OpenAI 兼容的接口格式。
"""
from typing import TYPE_CHECKING
from .base import OpenAICompatAdapter, RequestData
if TYPE_CHECKING:
from ..service import LLMModel
class ZhipuAdapter(OpenAICompatAdapter):
"""智谱AI适配器 - 使用智谱AI专用的OpenAI兼容接口"""
@property
def api_type(self) -> str:
return "zhipu"
@property
def supported_api_types(self) -> list[str]:
return ["zhipu"]
def get_chat_endpoint(self) -> str:
"""返回智谱AI聊天完成端点"""
return "/api/paas/v4/chat/completions"
def get_embedding_endpoint(self) -> str:
"""返回智谱AI嵌入端点"""
return "/v4/embeddings"
def prepare_simple_request(
self,
model: "LLMModel",
api_key: str,
prompt: str,
history: list[dict[str, str]] | None = None,
) -> RequestData:
"""准备简单文本生成请求 - 智谱AI优化实现"""
url = self.get_api_url(model, self.get_chat_endpoint())
headers = self.get_base_headers(api_key)
messages = []
if history:
messages.extend(history)
messages.append({"role": "user", "content": prompt})
body = {
"model": model.model_name,
"messages": messages,
}
body = self.apply_config_override(model, body)
return RequestData(url=url, headers=headers, body=body)