flowllm 0.1.0__py3-none-any.whl → 0.1.2__py3-none-any.whl

llmflow/schema/message.py

@@ -1,68 +0,0 @@
-import json
-from typing import List
-
-from pydantic import BaseModel, Field, model_validator
-
-from llmflow.enumeration.role import Role
-
-
-class ToolCall(BaseModel):
-    index: int = Field(default=0)
-    id: str = Field(default="")
-    name: str = Field(default="")
-    arguments: str = Field(default="")
-    type: str = Field(default="function")
-
-    @model_validator(mode="before")  # noqa
-    @classmethod
-    def init_tool_call(cls, data: dict):
-        tool_type = data.get("type", "")
-        tool_type_dict = data.get(tool_type, {})
-
-        for key in ["name", "arguments"]:
-            if key not in data:
-                data[key] = tool_type_dict.get(key, "")
-        return data
-
-    @property
-    def argument_dict(self) -> dict:
-        return json.loads(self.arguments)
-
-    def simple_dump(self) -> dict:
-        return {
-            "id": self.id,
-            self.type: {
-                "arguments": self.arguments,
-                "name": self.name
-            },
-            "type": self.type,
-            "index": self.index,
-        }
-
-class Message(BaseModel):
-    role: Role = Field(default=Role.USER)
-    content: str | bytes = Field(default="")
-    reasoning_content: str = Field(default="")
-    tool_calls: List[ToolCall] = Field(default_factory=list)
-    tool_call_id: str = Field(default="")
-    metadata: dict = Field(default_factory=dict)
-
-    def simple_dump(self, add_reason_when_empty: bool = True) -> dict:
-        result: dict
-        if self.content:
-            result = {"role": self.role.value, "content": self.content}
-        elif add_reason_when_empty and self.reasoning_content:
-            result = {"role": self.role.value, "content": self.reasoning_content}
-        else:
-            result = {"role": self.role.value, "content": ""}
-
-        if self.tool_calls:
-            result["tool_calls"] = [x.simple_dump() for x in self.tool_calls]
-        return result
-
-
-class Trajectory(BaseModel):
-    task_id: str = Field(default="")
-    messages: List[Message] = Field(default_factory=list)
-    score: float = Field(default=0.0)
-    metadata: dict = Field(default_factory=dict)

llmflow/schema/request.py

@@ -1,32 +0,0 @@
-from typing import List
-
-from pydantic import BaseModel, Field
-
-from llmflow.schema.message import Message, Trajectory
-
-
-class BaseRequest(BaseModel):
-    workspace_id: str = Field(default="default")
-    config: dict = Field(default_factory=dict)
-
-
-class RetrieverRequest(BaseRequest):
-    query: str = Field(default="")
-    messages: List[Message] = Field(default_factory=list)
-    top_k: int = Field(default=1)
-
-
-class SummarizerRequest(BaseRequest):
-    traj_list: List[Trajectory] = Field(default_factory=list)
-
-
-class VectorStoreRequest(BaseRequest):
-    action: str = Field(default="")
-    src_workspace_id: str = Field(default="")
-    path: str = Field(default="")
-
-
-class AgentRequest(BaseRequest):
-    query: str = Field(default="")
-    messages: List[Message] = Field(default_factory=list)
-

llmflow/schema/response.py

@@ -1,29 +0,0 @@
-from typing import List
-
-from pydantic import BaseModel, Field
-
-from llmflow.schema.experience import BaseExperience
-from llmflow.schema.message import Message
-
-
-class BaseResponse(BaseModel):
-    success: bool = Field(default=True)
-    metadata: dict = Field(default_factory=dict)
-
-
-class RetrieverResponse(BaseResponse):
-    experience_list: List[BaseExperience] = Field(default_factory=list)
-    experience_merged: str = Field(default="")
-
-
-class SummarizerResponse(BaseResponse):
-    experience_list: List[BaseExperience] = Field(default_factory=list)
-    deleted_experience_ids: List[str] = Field(default_factory=list)
-
-
-class VectorStoreResponse(BaseResponse):
-    ...
-
-class AgentResponse(BaseResponse):
-    answer: str = Field(default="")
-    messages: List[Message] = Field(default_factory=list)

llmflow/service/llmflow_service.py

@@ -1,96 +0,0 @@
-from concurrent.futures import ThreadPoolExecutor
-from typing import List
-
-from loguru import logger
-
-from llmflow.config.config_parser import ConfigParser
-from llmflow.embedding_model import EMBEDDING_MODEL_REGISTRY
-from llmflow.pipeline.pipeline import Pipeline
-from llmflow.pipeline.pipeline_context import PipelineContext
-from llmflow.schema.app_config import AppConfig, HttpServiceConfig, EmbeddingModelConfig
-from llmflow.schema.request import SummarizerRequest, RetrieverRequest, VectorStoreRequest, AgentRequest, \
-    BaseRequest
-from llmflow.schema.response import SummarizerResponse, RetrieverResponse, VectorStoreResponse, AgentResponse, \
-    BaseResponse
-from llmflow.vector_store import VECTOR_STORE_REGISTRY
-
-
-class LLMFlowService:
-
-    def __init__(self, args: List[str]):
-        self.config_parser = ConfigParser(args)
-        self.init_app_config: AppConfig = self.config_parser.get_app_config()
-        self.thread_pool = ThreadPoolExecutor(max_workers=self.init_app_config.thread_pool.max_workers)
-
-        # The vectorstore is initialized at the very beginning and then used directly afterward.
-        self.vector_store_dict: dict = {}
-        for name, config in self.init_app_config.vector_store.items():
-            assert config.backend in VECTOR_STORE_REGISTRY, f"backend={config.backend} is not existed"
-            vector_store_cls = VECTOR_STORE_REGISTRY[config.backend]
-
-            assert config.embedding_model in self.init_app_config.embedding_model, \
-                f"embedding_model={config.embedding_model} is not existed"
-            embedding_model_config: EmbeddingModelConfig = self.init_app_config.embedding_model[config.embedding_model]
-
-            assert embedding_model_config.backend in EMBEDDING_MODEL_REGISTRY, \
-                f"embedding_model={embedding_model_config.backend} is not existed"
-            embedding_model_cls = EMBEDDING_MODEL_REGISTRY[embedding_model_config.backend]
-            embedding_model = embedding_model_cls(model_name=embedding_model_config.model_name,
-                                                  **embedding_model_config.params)
-
-            self.vector_store_dict[name] = vector_store_cls(embedding_model=embedding_model, **config.params)
-
-    @property
-    def http_service_config(self) -> HttpServiceConfig:
-        return self.init_app_config.http_service
-
-    def __call__(self, api: str, request: dict | BaseRequest) -> BaseResponse:
-        if isinstance(request, dict):
-            app_config: AppConfig = self.config_parser.get_app_config(**request["config"])
-        else:
-            app_config: AppConfig = self.config_parser.get_app_config(**request.config)
-
-        if api == "retriever":
-            if isinstance(request, dict):
-                request = RetrieverRequest(**request)
-            response = RetrieverResponse()
-            pipeline = app_config.api.retriever
-
-        elif api == "summarizer":
-            if isinstance(request, dict):
-                request = SummarizerRequest(**request)
-            response = SummarizerResponse()
-            pipeline = app_config.api.summarizer
-
-        elif api == "vector_store":
-            if isinstance(request, dict):
-                request = VectorStoreRequest(**request)
-            response = VectorStoreResponse()
-            pipeline = app_config.api.vector_store
-
-        elif api == "agent":
-            if isinstance(request, dict):
-                request = AgentRequest(**request)
-            response = AgentResponse()
-            pipeline = app_config.api.agent
-
-        else:
-            raise RuntimeError(f"Invalid service.api={api}")
-
-        logger.info(f"request={request.model_dump_json()}")
-
-        try:
-            context = PipelineContext(app_config=app_config,
-                                      thread_pool=self.thread_pool,
-                                      request=request,
-                                      response=response,
-                                      vector_store_dict=self.vector_store_dict)
-            pipeline = Pipeline(pipeline=pipeline, context=context)
-            pipeline()
-
-        except Exception as e:
-            logger.exception(f"api={api} encounter error={e.args}")
-            response.success = False
-            response.metadata["error"] = str(e)
-
-        return response

llmflow/tool/__init__.py

@@ -1,9 +0,0 @@
-from llmflow.utils.registry import Registry
-
-TOOL_REGISTRY = Registry()
-
-from llmflow.tool.code_tool import CodeTool
-from llmflow.tool.dashscope_search_tool import DashscopeSearchTool
-from llmflow.tool.tavily_search_tool import TavilySearchTool
-from llmflow.tool.terminate_tool import TerminateTool
-from llmflow.tool.mcp_tool import MCPTool

llmflow/tool/base_tool.py

@@ -1,80 +0,0 @@
-from abc import ABC
-
-from loguru import logger
-from pydantic import BaseModel, Field
-
-
-class BaseTool(BaseModel, ABC):
-    tool_id: str = Field(default="")
-    name: str = Field(..., description="tool name")
-    description: str = Field(..., description="tool description")
-    tool_type: str = Field(default="function")
-    parameters: dict = Field(default_factory=dict, description="tool parameters")
-    arguments: dict = Field(default_factory=dict, description="execute arguments")
-
-    enable_cache: bool = Field(default=False, description="whether to cache the tool result")
-    cached_result: dict = Field(default_factory=dict, description="tool execution result")
-
-    max_retries: int = Field(default=3, description="max retries")
-    raise_exception: bool = Field(default=True, description="raise exception")
-    success: bool = Field(default=True, description="whether the tool executed successfully")
-
-    def reset(self):
-        self.arguments.clear()
-        self.cached_result.clear()
-        self.success = True
-
-    def _execute(self, **kwargs):
-        raise NotImplementedError
-
-    def execute(self, **kwargs):
-        cache_id = ""
-        if self.enable_cache:
-            cache_id = self.get_cache_id(**kwargs)
-            if cache_id in self.cached_result:
-                return self.cached_result[cache_id]
-
-        for i in range(self.max_retries):
-            try:
-                if self.enable_cache:
-                    self.cached_result[cache_id] = self._execute(**kwargs)
-                    return self.cached_result[cache_id]
-
-                else:
-                    return self._execute(**kwargs)
-
-            except Exception as e:
-                logger.exception(f"using tool.name={self.name} encounter error with e={e.args}")
-                if i == self.max_retries - 1 and self.raise_exception:
-                    raise e
-
-        return None
-
-
-    def simple_dump(self) -> dict:
-        """
-        It may be in other different tool params formats; different versions are completed here.
-        """
-        return {
-            "type": self.tool_type,
-            self.tool_type: {
-                "name": self.name,
-                "description": self.description,
-                "parameters": self.parameters,
-            },
-        }
-
-    @property
-    def input_schema(self) -> dict:
-        return self.parameters.get("properties", {})
-
-    @property
-    def output_schema(self) -> dict:
-        raise NotImplementedError
-
-    def refresh(self):
-        # for mcp
-        raise NotImplementedError
-
-    def get_cache_id(self, **kwargs) -> str:
-        raise NotImplementedError

llmflow/tool/code_tool.py

@@ -1,43 +0,0 @@
-import sys
-from io import StringIO
-
-from llmflow.tool import TOOL_REGISTRY
-from llmflow.tool.base_tool import BaseTool
-
-
-@TOOL_REGISTRY.register()
-class CodeTool(BaseTool):
-    name: str = "python_execute"
-    description: str = "Execute python code can be used in scenarios such as analysis or calculation, and the final result can be printed using the `print` function."
-    parameters: dict = {
-        "type": "object",
-        "properties": {
-            "code": {
-                "type": "string",
-                "description": "code to be executed. Please do not execute any matplotlib code here.",
-            }
-        },
-        "required": ["code"]
-    }
-
-    def _execute(self, code: str, **kwargs):
-        old_stdout = sys.stdout
-        redirected_output = sys.stdout = StringIO()
-
-        try:
-            exec(code)
-            result = redirected_output.getvalue()
-
-        except Exception as e:
-            self.success = False
-            result = str(e)
-
-        sys.stdout = old_stdout
-
-        return result
-
-
-if __name__ == '__main__':
-    tool = CodeTool()
-    print(tool.execute(code="print('Hello World')"))
-    print(tool.execute(code="print('Hello World!'"))

llmflow/tool/dashscope_search_tool.py

@@ -1,162 +0,0 @@
-import os
-from typing import Literal
-
-import dashscope
-from dashscope.api_entities.dashscope_response import Message
-from dotenv import load_dotenv
-from loguru import logger
-from pydantic import Field
-
-from llmflow.tool import TOOL_REGISTRY
-from llmflow.tool.base_tool import BaseTool
-
-
-@TOOL_REGISTRY.register()
-class DashscopeSearchTool(BaseTool):
-    name: str = "web_search"
-    description: str = "Use search keywords to retrieve relevant information from the internet. " \
-                       "If there are multiple search keywords, please use each keyword separately to call this tool."
-    parameters: dict = {
-        "type": "object",
-        "properties": {
-            "query": {
-                "type": "string",
-                "description": "search keyword",
-            }
-        },
-        "required": ["query"]
-    }
-
-    model_name: Literal["qwen-plus-2025-04-28", "qwq-plus-latest", "qwen-max-2025-01-25"] = \
-        Field(default="qwen-plus-2025-04-28")
-    api_key: str = Field(default_factory=lambda: os.environ["DASHSCOPE_API_KEY"])
-    stream_print: bool = Field(default=False)
-    temperature: float = Field(default=0.0000001)
-    use_role_prompt: bool = Field(default=True)
-    role_prompt: str = """
-# user's question
-{question}
-
-# task
-Extract the original content related to the user's question directly from the context, maintain accuracy, and avoid excessive processing.    """.strip()
-    return_only_content: bool = Field(default=True)
-
-    def parse_reasoning_response(self, response, result: dict):
-        is_answering = False
-        is_first_chunk = True
-
-        for chunk in response:
-            if is_first_chunk:
-                result["search_results"] = chunk.output.search_info["search_results"]
-
-                if self.stream_print:
-                    print("=" * 20 + "search result" + "=" * 20)
-                    for web in result["search_results"]:
-                        print(f"[{web['index']}]: [{web['title']}]({web['url']})")
-                    print("=" * 20 + "thinking process" + "=" * 20)
-                result["reasoning_content"] += chunk.output.choices[0].message.reasoning_content
-
-                if self.stream_print:
-                    print(chunk.output.choices[0].message.reasoning_content, end="", flush=True)
-                is_first_chunk = False
-
-            else:
-                if chunk.output.choices[0].message.content == "" \
-                        and chunk.output.choices[0].message.reasoning_content == "":
-                    pass
-
-                else:
-                    if chunk.output.choices[0].message.reasoning_content != "" and \
-                            chunk.output.choices[0].message.content == "":
-
-                        if self.stream_print:
-                            print(chunk.output.choices[0].message.reasoning_content, end="", flush=True)
-                        result["reasoning_content"] += chunk.output.choices[0].message.reasoning_content
-
-                    elif chunk.output.choices[0].message.content != "":
-                        if not is_answering:
-                            if self.stream_print:
-                                print("\n" + "=" * 20 + "complete answer" + "=" * 20)
-                            is_answering = True
-
-                        if self.stream_print:
-                            print(chunk.output.choices[0].message.content, end="", flush=True)
-                        result["answer_content"] += chunk.output.choices[0].message.content
-
-    def parse_response(self, response, result: dict):
-        is_first_chunk = True
-
-        for chunk in response:
-            if is_first_chunk:
-                result["search_results"] = chunk.output.search_info["search_results"]
-
-                if self.stream_print:
-                    print("=" * 20 + "search result" + "=" * 20)
-                    for web in result["search_results"]:
-                        print(f"[{web['index']}]: [{web['title']}]({web['url']})")
-                    print("\n" + "=" * 20 + "complete answer" + "=" * 20)
-                is_first_chunk = False
-
-            else:
-                if chunk.output.choices[0].message.content == "":
-                    pass
-
-                else:
-                    if chunk.output.choices[0].message.content != "":
-                        if self.stream_print:
-                            print(chunk.output.choices[0].message.content, end="", flush=True)
-                        result["answer_content"] += chunk.output.choices[0].message.content
-
-    def execute(self, query: str = "", **kwargs):
-        result = {
-            "search_results": [],
-            "reasoning_content": "",
-            "answer_content": ""
-        }
-        user_query = self.role_prompt.format(question=query) if self.use_role_prompt else query
-        messages = [Message(role="user", content=user_query)]
-
-        response = dashscope.Generation.call(
-            api_key=self.api_key,
-            model=self.model_name,
-            messages=messages,
-            enable_thinking=True,
-            enable_search=True,
-            search_options={
-                "forced_search": True,
-                "enable_source": True,
-                "enable_citation": False,
-                "search_strategy": "pro"
-            },
-            stream=True,
-            incremental_output=True,
-            result_format="message",
-        )
-
-        if self.model_name != "qwen-max-2025-01-25":
-            self.parse_reasoning_response(response, result)
-        else:
-            self.parse_response(response, result)
-
-        if self.return_only_content:
-            return result["answer_content"]
-        else:
-            return result
-
-
-def main():
-    load_dotenv()
-    query = "What is artificial intelligence?"
-
-    tool = DashscopeSearchTool(stream_print=True)
-    logger.info(tool.execute(query=query))
-
-    tool = DashscopeSearchTool(stream_print=False)
-    logger.info(tool.execute(query=query))
-
-    tool = DashscopeSearchTool(stream_print=True, model_name="qwen-max-2025-01-25")
-    logger.info(tool.execute(query=query))
-
-
-if __name__ == '__main__':
-    main()

llmflow/tool/mcp_tool.py

@@ -1,77 +0,0 @@
-import asyncio
-from typing import List
-
-from mcp import ClientSession
-from mcp.client.sse import sse_client
-from pydantic import Field, model_validator
-
-from llmflow.tool import TOOL_REGISTRY
-from llmflow.tool.base_tool import BaseTool
-
-
-@TOOL_REGISTRY.register()
-class MCPTool(BaseTool):
-    server_url: str = Field(..., description="MCP server URL")
-    tool_name_list: List[str] = Field(default_factory=list)
-    cache_tools: dict = Field(default_factory=dict, alias="cache_tools")
-
-    @model_validator(mode="after")
-    def refresh_tools(self):
-        self.refresh()
-        return self
-
-    async def _get_tools(self):
-        async with sse_client(url=self.server_url) as streams:
-            async with ClientSession(streams[0], streams[1]) as session:
-                await session.initialize()
-                tools = await session.list_tools()
-        return tools
-
-    def refresh(self):
-        self.tool_name_list.clear()
-        self.cache_tools.clear()
-
-        if "sse" in self.server_url:
-            original_tool_list = asyncio.run(self._get_tools())
-            for tool in original_tool_list.tools:
-                self.cache_tools[tool.name] = tool
-                self.tool_name_list.append(tool.name)
-        else:
-            raise NotImplementedError("Non-SSE refresh not implemented yet")
-
-    @property
-    def input_schema(self) -> dict:
-        return {x: self.cache_tools[x].inputSchema for x in self.cache_tools}
-
-    @property
-    def output_schema(self) -> dict:
-        raise NotImplementedError("Output schema not implemented yet")
-
-    def get_tool_description(self, tool_name: str, schema: bool = False) -> str:
-        if tool_name not in self.cache_tools:
-            raise RuntimeError(f"Tool {tool_name} not found")
-
-        tool = self.cache_tools.get(tool_name)
-        description = f"tool={tool_name} description={tool.description}\n"
-        if schema:
-            description += f"input_schema={self.input_schema[tool_name]}\n" \
-                           f"output_schema={self.output_schema[tool_name]}\n"
-        return description.strip()
-
-    async def async_execute(self, tool_name: str, **kwargs):
-        if "sse" in self.server_url:
-            async with sse_client(url=self.server_url) as streams:
-                async with ClientSession(streams[0], streams[1]) as session:
-                    await session.initialize()
-                    results = await session.call_tool(tool_name, kwargs)
-            return results.content[0].text, results.isError
-
-        else:
-            raise NotImplementedError("Non-SSE execute not implemented yet")
-
-    def _execute(self, **kwargs):
-        return asyncio.run(self.async_execute(**kwargs))
-
-    def get_cache_id(self, **kwargs) -> str:
-        # Implement a method to generate a unique cache ID based on the input
-        return f"{kwargs.get('tool_name')}_{hash(frozenset(kwargs.get('args', {}).items()))}"