isa-model 0.4.0__py3-none-any.whl → 0.4.3__py3-none-any.whl

isa_model/inference/services/llm/huggingface_llm_service.py

@@ -0,0 +1,581 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+
+"""
+ISA LLM Service - Inference client for Modal-deployed HuggingFace models
+Supports custom trained models deployed on Modal infrastructure
+"""
+
+import logging
+import os
+from typing import Dict, Any, Optional, List
+
+try:
+    import modal
+    MODAL_AVAILABLE = True
+except ImportError:
+    MODAL_AVAILABLE = False
+    modal = None
+
+from isa_model.inference.services.base_service import BaseService
+from isa_model.core.models.model_manager import ModelManager
+from isa_model.core.config import ConfigManager
+
+logger = logging.getLogger(__name__)
+
+class ISALLMService(BaseService):
+    """
+    ISA LLM Service - Client for Modal-deployed HuggingFace models
+    Calls ISA's own deployed LLM inference services on Modal
+    """
+    
+    def __init__(
+        self,
+        provider_name: str = "isa",
+        model_name: str = None,
+        model_manager: ModelManager = None,
+        config_manager: ConfigManager = None,
+        modal_app_name: str = "isa-llm-inference",
+        timeout: int = 60,
+        **kwargs
+    ):
+        # Skip BaseService init to avoid config validation for now
+        self.provider_name = provider_name
+        self.model_name = model_name or "isa-llm-service"
+        self.modal_app_name = modal_app_name
+        self.timeout = timeout
+        
+        # Initialize Modal client
+        if MODAL_AVAILABLE:
+            try:
+                # Get deployed Modal app
+                self.modal_app = modal.App.lookup(modal_app_name)
+                logger.info(f"Connected to Modal LLM app: {modal_app_name}")
+                
+                self.modal_service = True
+                logger.info("Modal LLM service connection established")
+                    
+            except Exception as e:
+                logger.warning(f"Failed to connect to Modal LLM app: {e}")
+                self.modal_app = None
+                self.modal_service = None
+        else:
+            logger.warning("Modal SDK not available")
+            self.modal_app = None
+            self.modal_service = None
+        
+        # Service statistics
+        self.request_count = 0
+        self.total_cost = 0.0
+        
+        # Fallback mode for when Modal is not available
+        self.fallback_mode = not MODAL_AVAILABLE or not self.modal_service
+    
+    async def _fallback_response(self, method_name: str, **kwargs) -> Dict[str, Any]:
+        """
+        Provide fallback responses when Modal service is not available
+        """
+        import time
+        import random
+        
+        if method_name == "generate_text":
+            prompt = kwargs.get("prompt", "")
+            # Simple rule-based responses for demo purposes
+            responses = [
+                "这是一个模拟的ISA LLM响应。",
+                "抱歉，Modal服务当前不可用，这是一个fallback响应。",
+                "ISA模型正在维护中，请稍后再试。",
+                f"您说：{prompt}。我理解了，但当前模型不可用。"
+            ]
+            
+            generated_text = random.choice(responses)
+            
+            return {
+                "success": True,
+                "text": generated_text,
+                "full_text": prompt + " " + generated_text,
+                "prompt": prompt,
+                "model_id": kwargs.get("model_id", "isa-llm-fallback"),
+                "provider": "ISA",
+                "service": "isa-llm",
+                "fallback": True,
+                "generation_config": kwargs.get("generation_config", {}),
+                "metadata": {
+                    "processing_time": random.uniform(0.5, 2.0),
+                    "device": "cpu",
+                    "input_tokens": len(prompt.split()),
+                    "output_tokens": len(generated_text.split()),
+                    "note": "This is a fallback response - Modal service not available"
+                }
+            }
+            
+        elif method_name == "chat_completion":
+            messages = kwargs.get("messages", [])
+            user_message = ""
+            if messages:
+                user_message = messages[-1].get("content", "")
+            
+            chat_responses = [
+                "很抱歉，ISA模型当前不可用，这是一个模拟响应。",
+                "我是ISA模型的fallback版本，功能有限。",
+                f"我听到您说：{user_message}，但现在无法提供完整的回复。",
+                "Modal服务正在重启中，请稍后再试完整的ISA模型功能。"
+            ]
+            
+            response_text = random.choice(chat_responses)
+            
+            return {
+                "success": True,
+                "text": response_text,
+                "role": "assistant",
+                "messages": messages,
+                "model_id": kwargs.get("model_id", "isa-llm-fallback"),
+                "provider": "ISA",
+                "service": "isa-llm",
+                "fallback": True,
+                "metadata": {
+                    "processing_time": random.uniform(0.3, 1.5),
+                    "device": "cpu",
+                    "note": "This is a fallback response - Modal service not available"
+                }
+            }
+            
+        elif method_name == "get_model_info":
+            return {
+                "success": True,
+                "model_id": kwargs.get("model_id", "isa-llm-fallback"),
+                "provider": "ISA",
+                "service": "isa-llm",
+                "architecture": "unknown (fallback mode)",
+                "fallback": True,
+                "note": "Modal service not available - showing fallback info"
+            }
+            
+        elif method_name == "health_check":
+            return {
+                "success": True,
+                "status": "fallback",
+                "service": "isa-llm",
+                "provider": "ISA",
+                "device": "cpu",
+                "fallback": True,
+                "message": "Modal service not available - running in fallback mode"
+            }
+        
+        else:
+            return {
+                "success": False,
+                "error": f"Method {method_name} not supported in fallback mode",
+                "fallback": True
+            }
+    
+    async def _call_modal_llm_service(
+        self,
+        method_name: str,
+        **kwargs
+    ) -> Dict[str, Any]:
+        """
+        Call Modal LLM service via SDK with improved error handling and fallback
+        """
+        # If in fallback mode, use fallback response immediately
+        if self.fallback_mode:
+            logger.info(f"Using fallback mode for {method_name}")
+            return await self._fallback_response(method_name, **kwargs)
+        
+        try:
+            if not MODAL_AVAILABLE:
+                logger.warning("Modal SDK not available, switching to fallback mode")
+                self.fallback_mode = True
+                return await self._fallback_response(method_name, **kwargs)
+            
+            if not self.modal_app or not self.modal_service:
+                logger.warning("Modal app/service not available, switching to fallback mode")
+                self.fallback_mode = True
+                return await self._fallback_response(method_name, **kwargs)
+            
+            logger.info(f"Calling Modal LLM service method: {method_name}")
+            
+            try:
+                # Use Modal SDK to call the service
+                ISALLMServiceCls = modal.Cls.from_name(
+                    app_name=self.modal_app_name,
+                    name="ISALLMService"
+                )
+                
+                # Create instance and call method
+                instance = ISALLMServiceCls()
+                method = getattr(instance, method_name)
+                result = method.remote(**kwargs)
+                
+                logger.info("✅ Modal LLM service call successful")
+                return result
+                
+            except modal.exception.NotFoundError:
+                logger.warning(f"Modal app not found, switching to fallback mode")
+                self.fallback_mode = True
+                return await self._fallback_response(method_name, **kwargs)
+                
+            except modal.exception.ConnectionError:
+                logger.warning(f"Modal connection error, switching to fallback mode")
+                self.fallback_mode = True
+                return await self._fallback_response(method_name, **kwargs)
+                        
+        except Exception as e:
+            logger.error(f"Modal LLM service call failed: {e}, switching to fallback mode")
+            self.fallback_mode = True
+            return await self._fallback_response(method_name, **kwargs)
+    
+    async def complete(
+        self,
+        prompt: str,
+        model_id: str = None,
+        max_length: Optional[int] = 50,
+        temperature: float = 0.7,
+        do_sample: bool = True,
+        top_p: float = 0.9,
+        repetition_penalty: float = 1.1,
+        **kwargs
+    ) -> Dict[str, Any]:
+        """
+        Generate completion using Modal-deployed LLM service
+        
+        Args:
+            prompt: Input text prompt
+            model_id: HuggingFace model ID to use
+            max_length: Maximum length of generated text
+            temperature: Sampling temperature
+            do_sample: Whether to use sampling
+            top_p: Top-p sampling parameter
+            repetition_penalty: Repetition penalty
+            **kwargs: Additional generation parameters
+            
+        Returns:
+            Dictionary containing generated text and metadata
+        """
+        try:
+            # Get HF token from environment
+            hf_token = os.getenv("HF_TOKEN")
+            
+            # Use provided model_id or default trained model
+            target_model = model_id or "xenobordom/dialogpt-isa-trained-1755493402"
+            
+            # Call Modal service
+            result = await self._call_modal_llm_service(
+                method_name="generate_text",
+                prompt=prompt,
+                model_id=target_model,
+                hf_token=hf_token,
+                max_length=max_length,
+                temperature=temperature,
+                do_sample=do_sample,
+                top_p=top_p,
+                repetition_penalty=repetition_penalty,
+                **kwargs
+            )
+            
+            if result and result.get('success', False):
+                self.request_count += 1
+                
+                # Add cost tracking if available
+                if 'billing' in result:
+                    cost = result['billing'].get('estimated_cost_usd', 0)
+                    self.total_cost += cost
+                
+                return result
+            else:
+                return {
+                    'success': False,
+                    'provider': 'ISA',
+                    'service': 'isa-llm',
+                    'error': f'Modal LLM service returned error: {result.get("error", "Unknown error") if result else "No response"}',
+                    'details': result
+                }
+                
+        except Exception as e:
+            logger.error(f"ISA LLM completion failed: {e}")
+            return {
+                'success': False,
+                'provider': 'ISA',
+                'service': 'isa-llm',
+                'error': str(e)
+            }
+    
+    async def chat(
+        self,
+        messages: List[Dict[str, str]],
+        model_id: str = None,
+        **kwargs
+    ) -> Dict[str, Any]:
+        """
+        Chat completion using Modal-deployed LLM service
+        
+        Args:
+            messages: List of message dictionaries with 'role' and 'content'
+            model_id: HuggingFace model ID to use
+            **kwargs: Additional generation parameters
+            
+        Returns:
+            Dictionary containing generated response and metadata
+        """
+        try:
+            # Get HF token from environment
+            hf_token = os.getenv("HF_TOKEN")
+            
+            # Use provided model_id or default trained model
+            target_model = model_id or "xenobordom/dialogpt-isa-trained-1755493402"
+            
+            # Call Modal service
+            result = await self._call_modal_llm_service(
+                method_name="chat_completion",
+                messages=messages,
+                model_id=target_model,
+                hf_token=hf_token,
+                **kwargs
+            )
+            
+            if result and result.get('success', False):
+                self.request_count += 1
+                
+                # Add cost tracking if available
+                if 'billing' in result:
+                    cost = result['billing'].get('estimated_cost_usd', 0)
+                    self.total_cost += cost
+                
+                return result
+            else:
+                return {
+                    'success': False,
+                    'provider': 'ISA',
+                    'service': 'isa-llm',
+                    'error': f'Modal LLM service returned error: {result.get("error", "Unknown error") if result else "No response"}',
+                    'details': result
+                }
+                
+        except Exception as e:
+            logger.error(f"ISA LLM chat completion failed: {e}")
+            return {
+                'success': False,
+                'provider': 'ISA',
+                'service': 'isa-llm',
+                'error': str(e)
+            }
+    
+    async def get_model_info(self, model_id: str = None) -> Dict[str, Any]:
+        """Get information about the model via Modal service"""
+        try:
+            # Get HF token from environment
+            hf_token = os.getenv("HF_TOKEN")
+            
+            # Use provided model_id or default trained model
+            target_model = model_id or "xenobordom/dialogpt-isa-trained-1755493402"
+            
+            # Call Modal service
+            result = await self._call_modal_llm_service(
+                method_name="get_model_info",
+                model_id=target_model,
+                hf_token=hf_token
+            )
+            
+            if result and result.get('success', False):
+                return result
+            else:
+                return {
+                    'success': False,
+                    'provider': 'ISA',
+                    'service': 'isa-llm',
+                    'error': f'Modal LLM service returned error: {result.get("error", "Unknown error") if result else "No response"}'
+                }
+                
+        except Exception as e:
+            logger.error(f"Error getting model info: {e}")
+            return {
+                'success': False,
+                'error': str(e)
+            }
+    
+    async def health_check(self) -> Dict[str, Any]:
+        """Check ISA LLM service health"""
+        try:
+            # Call Modal service health check
+            result = await self._call_modal_llm_service(
+                method_name="health_check"
+            )
+            
+            if result and result.get('success', False):
+                return {
+                    'success': True,
+                    'provider': 'ISA',
+                    'service': 'isa-llm',
+                    'status': 'healthy',
+                    'modal_service': result,
+                    'usage_stats': {
+                        'total_requests': self.request_count,
+                        'total_cost_usd': round(self.total_cost, 6)
+                    }
+                }
+            else:
+                return {
+                    'success': False,
+                    'provider': 'ISA',
+                    'service': 'isa-llm',
+                    'status': 'error',
+                    'error': f'Modal service error: {result.get("error", "Unknown error") if result else "No response"}'
+                }
+                
+        except Exception as e:
+            return {
+                'success': False,
+                'provider': 'ISA',
+                'service': 'isa-llm',
+                'status': 'error',
+                'error': str(e)
+            }
+    
+    def get_supported_tasks(self) -> List[str]:
+        """Get supported task list"""
+        return [
+            'generate',     # Text generation
+            'chat',         # Chat completion
+            'complete'      # Text completion
+        ]
+    
+    def get_supported_models(self) -> List[str]:
+        """Get supported model types"""
+        return [
+            'dialogpt',     # DialoGPT models
+            'gpt2',         # GPT-2 models
+            'custom'        # Custom trained models
+        ]
+    
+    async def invoke(self, input_data: str, task: str = "chat", **kwargs) -> Dict[str, Any]:
+        """
+        Unified invoke method for ISA LLM service compatibility
+        Required by the ISA Model client interface
+        """
+        try:
+            if task in ["chat", "generate", "complete"]:
+                # Handle chat tasks by converting to message format
+                if task == "chat":
+                    if isinstance(input_data, str):
+                        messages = [{"role": "user", "content": input_data}]
+                    elif isinstance(input_data, list):
+                        messages = input_data
+                    else:
+                        messages = [{"role": "user", "content": str(input_data)}]
+                    
+                    result = await self.chat(messages, **kwargs)
+                    
+                    # Convert result to unified format
+                    if result.get('success'):
+                        response_text = ""
+                        if 'response' in result and isinstance(result['response'], dict):
+                            response_text = result['response'].get('generated_text', '')
+                        elif 'generated_text' in result:
+                            response_text = result['generated_text']
+                        elif 'content' in result:
+                            response_text = result['content']
+                        
+                        return {
+                            'success': True,
+                            'result': {
+                                'content': response_text,
+                                'tool_calls': [],
+                                'response_metadata': result.get('metadata', {})
+                            },
+                            'error': None,
+                            'metadata': {
+                                'model_used': self.model_name,
+                                'provider': self.provider_name,
+                                'task': task,
+                                'service_type': 'text',
+                                'processing_time': result.get('processing_time', 0)
+                            }
+                        }
+                    else:
+                        return {
+                            'success': False,
+                            'result': None,
+                            'error': result.get('error', 'Unknown error'),
+                            'metadata': {
+                                'model_used': self.model_name,
+                                'provider': self.provider_name,
+                                'task': task,
+                                'service_type': 'text'
+                            }
+                        }
+                        
+                elif task in ["generate", "complete"]:
+                    result = await self.complete(input_data, **kwargs)
+                    
+                    # Convert result to unified format
+                    if result.get('success'):
+                        response_text = ""
+                        if 'response' in result and isinstance(result['response'], dict):
+                            response_text = result['response'].get('generated_text', '')
+                        elif 'generated_text' in result:
+                            response_text = result['generated_text']
+                        elif 'content' in result:
+                            response_text = result['content']
+                        
+                        return {
+                            'success': True,
+                            'result': {
+                                'content': response_text,
+                                'response_metadata': result.get('metadata', {})
+                            },
+                            'error': None,
+                            'metadata': {
+                                'model_used': self.model_name,
+                                'provider': self.provider_name,
+                                'task': task,
+                                'service_type': 'text',
+                                'processing_time': result.get('processing_time', 0)
+                            }
+                        }
+                    else:
+                        return {
+                            'success': False,
+                            'result': None,
+                            'error': result.get('error', 'Unknown error'),
+                            'metadata': {
+                                'model_used': self.model_name,
+                                'provider': self.provider_name,
+                                'task': task,
+                                'service_type': 'text'
+                            }
+                        }
+            else:
+                return {
+                    'success': False,
+                    'result': None,
+                    'error': f'Unsupported task: {task}. Supported tasks: {self.get_supported_tasks()}',
+                    'metadata': {
+                        'model_used': self.model_name,
+                        'provider': self.provider_name,
+                        'task': task,
+                        'service_type': 'text'
+                    }
+                }
+                
+        except Exception as e:
+            logger.error(f"ISA LLM invoke failed: {e}")
+            return {
+                'success': False,
+                'result': None,
+                'error': str(e),
+                'metadata': {
+                    'model_used': self.model_name,
+                    'provider': self.provider_name,
+                    'task': task,
+                    'service_type': 'text'
+                }
+            }
+
+# Backward compatibility aliases
+class HuggingFaceLLMService(ISALLMService):
+    """Alias for backward compatibility with AIFactory naming convention"""
+    pass
+
+class HuggingFaceInferenceService(ISALLMService):
+    """Alias for backward compatibility"""
+    pass