isa-model 0.1.0__py3-none-any.whl → 0.1.1__py3-none-any.whl

isa_model/inference/services/vision/replicate_vision_service.py

@@ -1,241 +0,0 @@
-#!/usr/bin/env python
-# -*- coding: utf-8 -*-
-
-"""
-Replicate Vision服务
-用于与Replicate API交互，支持图像生成和图像分析
-"""
-
-import os
-import time
-import uuid
-import logging
-from typing import Dict, Any, List, Optional, Union
-import asyncio
-import aiohttp
-import replicate
-import requests
-from PIL import Image
-from io import BytesIO
-
-from isa_model.inference.services.base_service import BaseService
-from isa_model.inference.providers.base_provider import BaseProvider
-from isa_model.inference.base import ModelType
-
-# 设置日志记录
-logging.basicConfig(level=logging.INFO)
-logger = logging.getLogger(__name__)
-
-class ReplicateVisionService(BaseService):
-    """
-    Replicate Vision服务，用于处理图像生成和分析
-    """
-    
-    def __init__(self, provider: BaseProvider, model_name: str):
-        """
-        初始化Replicate Vision服务
-        
-        Args:
-            provider: Replicate提供商实例
-            model_name: Replicate模型ID (格式: 'username/model_name:version')
-        """
-        super().__init__(provider, model_name)
-        self.api_token = provider.config.get("api_token", os.environ.get("REPLICATE_API_TOKEN"))
-        self.client = replicate.Client(api_token=self.api_token)
-        self.model_type = ModelType.VISION
-        
-        # 可选的默认配置
-        self.guidance_scale = provider.config.get("guidance_scale", 7.5)
-        self.num_inference_steps = provider.config.get("num_inference_steps", 30)
-        
-        # 生成的图像存储目录
-        self.output_dir = "generated_images"
-        os.makedirs(self.output_dir, exist_ok=True)
-        
-    async def load(self) -> None:
-        """
-        加载模型（对于Replicate，这只是验证API令牌）
-        """
-        if not self.api_token:
-            raise ValueError("缺少Replicate API令牌，请设置REPLICATE_API_TOKEN环境变量")
-            
-        # 验证令牌有效性
-        try:
-            self.client.api_token = self.api_token
-            logger.info(f"Replicate Vision服务初始化成功，使用模型: {self.model_name}")
-        except Exception as e:
-            logger.error(f"Replicate初始化失败: {e}")
-            raise
-    
-    async def generate_image(self, input_data: Dict[str, Any]) -> Dict[str, Any]:
-        """
-        使用Replicate模型生成图像
-        
-        Args:
-            input_data: 包含生成参数的字典
-            
-        Returns:
-            包含生成图像URL的结果字典
-        """
-        try:
-            # 设置默认参数
-            if "guidance_scale" not in input_data and self.guidance_scale:
-                input_data["guidance_scale"] = self.guidance_scale
-                
-            if "num_inference_steps" not in input_data and self.num_inference_steps:
-                input_data["num_inference_steps"] = self.num_inference_steps
-            
-            # 运行模型（同步API调用）
-            logger.info(f"开始使用模型 {self.model_name} 生成图像")
-            
-            # 转换成异步操作
-            loop = asyncio.get_event_loop()
-            output = await loop.run_in_executor(
-                None, 
-                lambda: replicate.run(self.model_name, input=input_data)
-            )
-            
-            # 将结果转换为标准格式
-            # 处理Replicate对象输出
-            if hasattr(output, 'url'):
-                urls = [output.url]
-            elif isinstance(output, list) and all(hasattr(item, 'url') for item in output if item is not None):
-                urls = [item.url for item in output if item is not None]
-            else:
-                # 兼容直接返回URL字符串的情况
-                urls = output if isinstance(output, list) else [output]
-            
-            result = {
-                "urls": urls,
-                "metadata": {
-                    "model": self.model_name,
-                    "input": input_data
-                }
-            }
-            
-            logger.info(f"图像生成完成: {result['urls']}")
-            return result
-            
-        except Exception as e:
-            logger.error(f"图像生成失败: {e}")
-            raise
-    
-    async def analyze_image(self, image_path: str, prompt: str) -> Dict[str, Any]:
-        """
-        分析图像（用于支持视觉分析模型）
-        
-        Args:
-            image_path: 图像路径或URL
-            prompt: 分析提示
-            
-        Returns:
-            分析结果字典
-        """
-        try:
-            # 构建输入数据
-            input_data = {
-                "image": self._get_image_url(image_path),
-                "prompt": prompt
-            }
-            
-            # 运行模型
-            logger.info(f"开始使用模型 {self.model_name} 分析图像")
-            
-            # 转换成异步操作
-            loop = asyncio.get_event_loop()
-            output = await loop.run_in_executor(
-                None, 
-                lambda: replicate.run(self.model_name, input=input_data)
-            )
-            
-            result = {
-                "text": output,
-                "metadata": {
-                    "model": self.model_name,
-                    "input": input_data
-                }
-            }
-            
-            logger.info(f"图像分析完成")
-            return result
-            
-        except Exception as e:
-            logger.error(f"图像分析失败: {e}")
-            raise
-    
-    async def generate_and_save(self, input_data: Dict[str, Any]) -> Dict[str, Any]:
-        """
-        生成图像并保存到本地
-        
-        Args:
-            input_data: 包含生成参数的字典
-            
-        Returns:
-            包含生成图像URL和保存路径的结果字典
-        """
-        # 首先生成图像
-        result = await self.generate_image(input_data)
-        
-        # 然后下载并保存
-        saved_paths = []
-        for i, url in enumerate(result["urls"]):
-            # 生成唯一文件名
-            timestamp = int(time.time())
-            file_name = f"{self.output_dir}/{timestamp}_{uuid.uuid4().hex[:8]}_{i+1}.png"
-            
-            # 异步下载图像
-            try:
-                await self._download_image(url, file_name)
-                saved_paths.append(file_name)
-                logger.info(f"图像已保存至: {file_name}")
-            except Exception as e:
-                logger.error(f"保存图像失败: {e}")
-        
-        # 添加保存路径到结果
-        result["saved_paths"] = saved_paths
-        return result
-    
-    async def _download_image(self, url: str, save_path: str) -> None:
-        """
-        异步下载图像并保存
-        
-        Args:
-            url: 图像URL
-            save_path: 保存路径
-        """
-        try:
-            async with aiohttp.ClientSession() as session:
-                async with session.get(url) as response:
-                    if response.status == 200:
-                        content = await response.read()
-                        img = Image.open(BytesIO(content))
-                        img.save(save_path)
-                    else:
-                        logger.error(f"下载图像失败: HTTP {response.status}")
-                        raise Exception(f"下载图像失败: HTTP {response.status}")
-        except Exception as e:
-            logger.error(f"下载图像时出错: {e}")
-            raise
-    
-    def _get_image_url(self, image_path: str) -> str:
-        """
-        获取图像URL（如果提供的是本地路径，则上传到临时存储）
-        
-        Args:
-            image_path: 图像路径或URL
-            
-        Returns:
-            图像URL
-        """
-        # 如果已经是URL，直接返回
-        if image_path.startswith(("http://", "https://")):
-            return image_path
-        
-        # 否则，这是一个需要上传的本地文件
-        # 注意：这里可以实现上传逻辑，但为简单起见，我们仅支持URL
-        raise NotImplementedError("当前仅支持图像URL，不支持上传本地文件")
-    
-    async def unload(self) -> None:
-        """卸载模型（对于Replicate API，这是一个无操作）"""
-        logger.info(f"卸载Replicate Vision服务: {self.model_name}")
-        # 没有需要清理的资源 

isa_model/inference/services/vision/triton_vision_service.py

@@ -1,199 +0,0 @@
-import json
-import logging
-import asyncio
-import base64
-import io
-from PIL import Image
-import numpy as np
-from typing import Dict, List, Any, AsyncGenerator, Optional, Union
-
-from isa_model.inference.services.base_service import BaseService
-from isa_model.inference.providers.triton_provider import TritonProvider
-
-logger = logging.getLogger(__name__)
-
-
-class TritonVisionService(BaseService):
-    """
-    Vision service that uses Triton Inference Server to run inference.
-    """
-    
-    def __init__(self, provider: TritonProvider, model_name: str):
-        """
-        Initialize the Triton Vision service.
-        
-        Args:
-            provider: The Triton provider
-            model_name: Name of the model in Triton (e.g., "Gemma3-4B")
-        """
-        super().__init__(provider, model_name)
-        self.client = None
-        self.token_usage = {"prompt_tokens": 0, "completion_tokens": 0, "total_tokens": 0}
-        self.last_token_usage = {"prompt_tokens": 0, "completion_tokens": 0, "total_tokens": 0}
-        
-    async def _initialize_client(self):
-        """Initialize the Triton client"""
-        if self.client is None:
-            self.client = self.provider.create_client()
-            
-            # Check if model is ready
-            if not self.provider.is_model_ready(self.model_name):
-                logger.error(f"Model {self.model_name} is not ready on Triton server")
-                raise RuntimeError(f"Model {self.model_name} is not ready on Triton server")
-                
-            logger.info(f"Initialized Triton client for vision model: {self.model_name}")
-    
-    async def process_image(self, 
-                          image: Union[str, Image.Image, bytes], 
-                          prompt: Optional[str] = None,
-                          params: Optional[Dict[str, Any]] = None) -> str:
-        """
-        Process an image and generate a description.
-        
-        Args:
-            image: Input image (PIL Image, base64 string, or bytes)
-            prompt: Optional text prompt to guide the model
-            params: Generation parameters
-            
-        Returns:
-            Generated text description
-        """
-        await self._initialize_client()
-        
-        try:
-            import tritonclient.http as httpclient
-            
-            # Process the image to get numpy array
-            image_array = self._prepare_image_input(image)
-            
-            # Create input tensors for the image
-            image_input = httpclient.InferInput("IMAGE", image_array.shape, "UINT8")
-            image_input.set_data_from_numpy(image_array)
-            inputs = [image_input]
-            
-            # Add text prompt if provided
-            if prompt:
-                text_data = np.array([prompt], dtype=np.object_)
-                text_input = httpclient.InferInput("TEXT", text_data.shape, "BYTES")
-                text_input.set_data_from_numpy(text_data)
-                inputs.append(text_input)
-            
-            # Add parameters if provided
-            if params:
-                default_params = {
-                    "max_new_tokens": 512,
-                    "temperature": 0.7,
-                    "top_p": 0.9,
-                    "do_sample": True
-                }
-                generation_params = {**default_params, **params}
-                
-                param_json = json.dumps(generation_params)
-                param_data = np.array([param_json], dtype=np.object_)
-                param_input = httpclient.InferInput("PARAMETERS", param_data.shape, "BYTES")
-                param_input.set_data_from_numpy(param_data)
-                inputs.append(param_input)
-            
-            # Create output tensor
-            outputs = [httpclient.InferRequestedOutput("TEXT")]
-            
-            # Send the request
-            response = await asyncio.to_thread(
-                self.client.infer,
-                self.model_name,
-                inputs,
-                outputs=outputs
-            )
-            
-            # Process the response
-            output = response.as_numpy("TEXT")
-            response_text = output[0].decode('utf-8')
-            
-            # Update token usage (estimated since we don't have actual token counts)
-            prompt_tokens = len(prompt) // 4 if prompt else 100  # Rough estimate
-            completion_tokens = len(response_text) // 4  # Rough estimate
-            total_tokens = prompt_tokens + completion_tokens
-            
-            self.last_token_usage = {
-                "prompt_tokens": prompt_tokens,
-                "completion_tokens": completion_tokens,
-                "total_tokens": total_tokens
-            }
-            
-            # Update total token usage
-            self.token_usage["prompt_tokens"] += prompt_tokens
-            self.token_usage["completion_tokens"] += completion_tokens
-            self.token_usage["total_tokens"] += total_tokens
-            
-            return response_text
-            
-        except Exception as e:
-            logger.error(f"Error during Triton vision inference: {str(e)}")
-            raise
-    
-    def get_token_usage(self) -> Dict[str, int]:
-        """
-        Get total token usage statistics.
-        
-        Returns:
-            Dictionary with token usage statistics
-        """
-        return self.token_usage
-    
-    def get_last_token_usage(self) -> Dict[str, int]:
-        """
-        Get token usage from last request.
-        
-        Returns:
-            Dictionary with token usage statistics from last request
-        """
-        return self.last_token_usage
-    
-    def _prepare_image_input(self, image: Union[str, Image.Image, bytes]) -> np.ndarray:
-        """
-        Process different types of image inputs into a numpy array.
-        
-        Args:
-            image: Image input (PIL Image, base64 string, or bytes)
-            
-        Returns:
-            Numpy array of the image
-        """
-        # Convert to PIL image first
-        pil_image = self._to_pil_image(image)
-        
-        # Convert PIL image to numpy array
-        return np.array(pil_image)
-    
-    def _to_pil_image(self, image: Union[str, Image.Image, bytes]) -> Image.Image:
-        """
-        Convert different image inputs to PIL Image.
-        
-        Args:
-            image: Image input (PIL Image, base64 string, or bytes)
-            
-        Returns:
-            PIL Image
-        """
-        if isinstance(image, Image.Image):
-            return image
-        
-        elif isinstance(image, str):
-            # Check if it's a base64 string
-            if image.startswith("data:image"):
-                # Extract the base64 part
-                image = image.split(",")[1]
-            
-            try:
-                # Try to decode as base64
-                image_bytes = base64.b64decode(image)
-                return Image.open(io.BytesIO(image_bytes))
-            except Exception:
-                # Try to open as a file path
-                return Image.open(image)
-        
-        elif isinstance(image, bytes):
-            return Image.open(io.BytesIO(image))
-        
-        else:
-            raise ValueError(f"Unsupported image type: {type(image)}") 

isa_model-0.1.0.dist-info/METADATA

@@ -1,116 +0,0 @@
-Metadata-Version: 2.4
-Name: isa-model
-Version: 0.1.0
-Summary: Unified AI model serving framework
-Author-email: isA_Model Contributors <your.email@example.com>
-License: MIT
-Classifier: Development Status :: 3 - Alpha
-Classifier: Intended Audience :: Developers
-Classifier: Operating System :: OS Independent
-Classifier: Programming Language :: Python :: 3
-Classifier: License :: OSI Approved :: MIT License
-Requires-Python: >=3.8
-Description-Content-Type: text/markdown
-License-File: LICENSE
-Requires-Dist: fastapi>=0.95.0
-Requires-Dist: numpy>=1.20.0
-Requires-Dist: httpx>=0.23.0
-Requires-Dist: pydantic>=2.0.0
-Requires-Dist: uvicorn>=0.22.0
-Requires-Dist: requests>=2.28.0
-Requires-Dist: aiohttp>=3.8.0
-Requires-Dist: transformers>=4.30.0
-Requires-Dist: langchain-core>=0.1.0
-Requires-Dist: tritonclient[grpc,http]>=2.30.0
-Requires-Dist: huggingface-hub>=0.16.0
-Requires-Dist: kubernetes>=25.3.0
-Requires-Dist: mlflow>=2.4.0
-Requires-Dist: torch>=2.0.0
-Dynamic: license-file
-
-# isA_Model - AI服务工厂
-
-isA_Model是一个轻量级AI服务工厂，用于统一管理和调用不同的AI模型和服务提供商。
-
-## 特性
-
-- 支持多种AI提供商(Ollama, OpenAI, Replicate, Triton)
-- 统一的API接口
-- 灵活的工厂模式
-- 异步支持
-- 单例模式，高效缓存
-
-## 安装
-
-```bash
-pip install -r requirements.txt
-```
-
-## 快速开始
-
-使用AI工厂很简单：
-
-```python
-from isa_model.inference.ai_factory import AIFactory
-from isa_model.inference.base import ModelType
-
-# 获取工厂实例
-factory = AIFactory()
-
-# LLM示例 - 使用Ollama
-llm = factory.get_llm(model_name="llama3.1", provider="ollama")
-response = await llm.generate("你好，请介绍一下自己。")
-print(response)
-
-# 图像生成示例 - 使用Replicate
-vision_service = factory.get_vision_model(
-    model_name="stability-ai/sdxl:c221b2b8ef527988fb59bf24a8b97c4561f1c671f73bd389f866bfb27c061316",
-    provider="replicate",
-    config={"api_token": "your_replicate_token"}
-)
-result = await vision_service.generate_image({
-    "prompt": "A beautiful sunset over mountains",
-    "num_inference_steps": 25
-})
-print(result["urls"])
-```
-
-## 工厂架构
-
-isA_Model使用三层架构:
-
-1. **客户端层** - 应用程序代码
-2. **服务层** - 模型服务实现(LLM, 图像, 嵌入等)
-3. **提供商层** - 底层API集成(Ollama, OpenAI, Replicate等)
-
-### 主要组件
-
-- `AIFactory` - 中央工厂类，提供模型和服务访问
-- `BaseService` - 所有服务的基类
-- `BaseProvider` - 所有提供商的基类
-- 特定服务实现 - 如`ReplicateVisionService`, `OllamaLLMService`等
-
-## 支持的模型类型
-
-- **LLM** - 大语言模型 
-- **VISION** - 图像生成和分析
-- **EMBEDDING** - 文本嵌入
-- **AUDIO** - 语音识别
-- **RERANK** - 重排序
-
-## 示例
-
-查看`test_*.py`文件获取更多使用示例。
-
-## 环境变量
-
-将API密钥和其他配置添加到`.env.local`文件中:
-
-```
-OPENAI_API_KEY=your_openai_key
-REPLICATE_API_TOKEN=your_replicate_token
-```
-
-## 许可证
-
-MIT