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

isa_model/inference/services/audio/base_stt_service.py

@@ -1,13 +1,172 @@
 from abc import ABC, abstractmethod
 from typing import Dict, Any, List, Union, Optional, BinaryIO
+import aiohttp
+import asyncio
+import tempfile
+import os
+import logging
+from io import BytesIO
 from isa_model.inference.services.base_service import BaseService
 
+logger = logging.getLogger(__name__)
+
 class BaseSTTService(BaseService):
-    """Base class for Speech-to-Text services with unified task dispatch"""
+    """Base class for Speech-to-Text services with unified task dispatch and URL support"""
+    
+    async def _prepare_audio_input(self, audio_input: Union[str, BinaryIO, bytes]) -> Union[str, BinaryIO]:
+        """
+        Prepare audio input by handling URLs, file paths, bytes data, and file objects
+        
+        Args:
+            audio_input: Audio input (URL, file path, bytes data, or file object)
+            
+        Returns:
+            Prepared audio input (local file path or file object)
+        """
+        if isinstance(audio_input, bytes):
+            # Handle bytes data from API uploads
+            logger.info(f"Converting bytes data to temporary file ({len(audio_input)} bytes)")
+            return await self._save_bytes_to_temp_file(audio_input)
+        elif isinstance(audio_input, str):
+            # Check if it's a URL
+            if audio_input.startswith(('http://', 'https://')):
+                logger.info(f"Downloading audio from URL: {audio_input}")
+                return await self._download_audio_url(audio_input)
+            else:
+                # Regular file path or base64 string
+                return audio_input
+        else:
+            # Already a file object
+            return audio_input
+    
+    async def _prepare_audio_input_with_context(self, audio_input: Union[str, BinaryIO, bytes], context: Dict[str, Any]) -> Union[str, BinaryIO]:
+        """
+        Prepare audio input with additional context from kwargs
+        
+        Args:
+            audio_input: Audio input (URL, file path, bytes data, or file object)
+            context: Additional context including filename, content_type
+            
+        Returns:
+            Prepared audio input (local file path or file object)
+        """
+        if isinstance(audio_input, bytes):
+            # Handle bytes data from API uploads
+            filename = context.get('filename')
+            content_type = context.get('content_type')
+            logger.info(f"Converting bytes data to temporary file ({len(audio_input)} bytes), filename={filename}, content_type={content_type}")
+            return await self._save_bytes_to_temp_file(audio_input, filename, content_type)
+        else:
+            return await self._prepare_audio_input(audio_input)
+    
+    async def _download_audio_url(self, url: str) -> str:
+        """
+        Download audio file from URL to temporary file
+        
+        Args:
+            url: HTTP/HTTPS URL to audio file
+            
+        Returns:
+            Path to downloaded temporary file
+            
+        Raises:
+            Exception: If download fails
+        """
+        try:
+            async with aiohttp.ClientSession() as session:
+                async with session.get(url) as response:
+                    if response.status != 200:
+                        raise Exception(f"Failed to download audio: HTTP {response.status}")
+                    
+                    # Get content type to determine file extension
+                    content_type = response.headers.get('Content-Type', '')
+                    file_ext = self._get_file_extension_from_content_type(content_type)
+                    
+                    # Create temporary file
+                    temp_file = tempfile.NamedTemporaryFile(
+                        delete=False, 
+                        suffix=file_ext,
+                        prefix='audio_download_'
+                    )
+                    
+                    # Download and save
+                    async for chunk in response.content.iter_chunked(8192):
+                        temp_file.write(chunk)
+                    
+                    temp_file.close()
+                    logger.info(f"Downloaded audio to temporary file: {temp_file.name}")
+                    return temp_file.name
+                    
+        except Exception as e:
+            logger.error(f"Failed to download audio from URL {url}: {e}")
+            raise Exception(f"Audio URL download failed: {e}") from e
+    
+    def _get_file_extension_from_content_type(self, content_type: str) -> str:
+        """Get appropriate file extension from Content-Type header"""
+        content_type_map = {
+            'audio/mpeg': '.mp3',
+            'audio/mp3': '.mp3',
+            'audio/wav': '.wav',
+            'audio/wave': '.wav',
+            'audio/x-wav': '.wav',
+            'audio/flac': '.flac',
+            'audio/ogg': '.ogg',
+            'audio/m4a': '.m4a',
+            'audio/mp4': '.mp4',
+            'audio/webm': '.webm'
+        }
+        return content_type_map.get(content_type.lower(), '.audio')
+    
+    async def _save_bytes_to_temp_file(self, audio_bytes: bytes, filename: Optional[str] = None, content_type: Optional[str] = None) -> str:
+        """
+        Save audio bytes data to temporary file
+        
+        Args:
+            audio_bytes: Audio data as bytes
+            filename: Optional filename to determine extension
+            content_type: Optional content type to determine extension
+            
+        Returns:
+            Path to temporary file containing audio data
+        """
+        try:
+            # Determine file extension from filename or content type
+            suffix = '.mp3'  # Default
+            if filename and '.' in filename:
+                suffix = '.' + filename.split('.')[-1]
+            elif content_type:
+                suffix = self._get_file_extension_from_content_type(content_type)
+            
+            # Create temporary file with proper audio extension
+            temp_file = tempfile.NamedTemporaryFile(
+                delete=False,
+                suffix=suffix,
+                prefix='audio_bytes_'
+            )
+            
+            # Write bytes data
+            temp_file.write(audio_bytes)
+            temp_file.close()
+            
+            logger.info(f"Saved {len(audio_bytes)} bytes to temporary file: {temp_file.name}")
+            return temp_file.name
+            
+        except Exception as e:
+            logger.error(f"Failed to save audio bytes to temporary file: {e}")
+            raise Exception(f"Audio bytes save failed: {e}") from e
+    
+    def _cleanup_temp_file(self, file_path: str):
+        """Clean up temporary downloaded file"""
+        try:
+            if file_path and file_path.startswith(tempfile.gettempdir()):
+                os.unlink(file_path)
+                logger.debug(f"Cleaned up temporary file: {file_path}")
+        except Exception as e:
+            logger.warning(f"Failed to cleanup temporary file {file_path}: {e}")
     
     async def invoke(
         self, 
-        audio_input: Union[str, BinaryIO, List[Union[str, BinaryIO]]],
+        audio_input: Union[str, BinaryIO, bytes, List[Union[str, BinaryIO, bytes]]],
         task: Optional[str] = None,
         **kwargs
     ) -> Union[Dict[str, Any], List[Dict[str, Any]]]:
@@ -30,33 +189,47 @@ class BaseSTTService(BaseService):
         # ==================== 语音转文本类任务 ====================
         if task == "transcribe":
             if isinstance(audio_input, list):
+                # Prepare all audio inputs (handle URLs)
+                prepared_inputs = []
+                for audio in audio_input:
+                    prepared_input = await self._prepare_audio_input_with_context(audio, kwargs)
+                    prepared_inputs.append(prepared_input)
                 return await self.transcribe_batch(
-                    audio_input,
+                    prepared_inputs,
                     kwargs.get("language"),
                     kwargs.get("prompt")
                 )
             else:
+                # Prepare single audio input (handle URLs)
+                prepared_input = await self._prepare_audio_input_with_context(audio_input, kwargs)
                 return await self.transcribe(
-                    audio_input,
+                    prepared_input,
                     kwargs.get("language"),
                     kwargs.get("prompt")
                 )
         elif task == "translate":
             if isinstance(audio_input, list):
                 raise ValueError("translate task requires single audio input")
-            return await self.translate(audio_input)
+            prepared_input = await self._prepare_audio_input_with_context(audio_input, kwargs)
+            return await self.translate(prepared_input)
         elif task == "batch_transcribe":
             if not isinstance(audio_input, list):
                 audio_input = [audio_input]
+            # Prepare all audio inputs (handle URLs)
+            prepared_inputs = []
+            for audio in audio_input:
+                prepared_input = await self._prepare_audio_input_with_context(audio, kwargs)
+                prepared_inputs.append(prepared_input)
             return await self.transcribe_batch(
-                audio_input,
+                prepared_inputs,
                 kwargs.get("language"),
                 kwargs.get("prompt")
             )
         elif task == "detect_language":
             if isinstance(audio_input, list):
                 raise ValueError("detect_language task requires single audio input")
-            return await self.detect_language(audio_input)
+            prepared_input = await self._prepare_audio_input_with_context(audio_input, kwargs)
+            return await self.detect_language(prepared_input)
         else:
             raise NotImplementedError(f"{self.__class__.__name__} does not support task: {task}")
     
@@ -72,7 +245,7 @@ class BaseSTTService(BaseService):
     @abstractmethod
     async def transcribe(
         self, 
-        audio_file: Union[str, BinaryIO], 
+        audio_file: Union[str, BinaryIO, bytes], 
         language: Optional[str] = None,
         prompt: Optional[str] = None
     ) -> Dict[str, Any]:
@@ -96,7 +269,7 @@ class BaseSTTService(BaseService):
     @abstractmethod
     async def translate(
         self, 
-        audio_file: Union[str, BinaryIO]
+        audio_file: Union[str, BinaryIO, bytes]
     ) -> Dict[str, Any]:
         """
         Translate audio file to English text
@@ -115,7 +288,7 @@ class BaseSTTService(BaseService):
     @abstractmethod
     async def transcribe_batch(
         self, 
-        audio_files: List[Union[str, BinaryIO]], 
+        audio_files: List[Union[str, BinaryIO, bytes]], 
         language: Optional[str] = None,
         prompt: Optional[str] = None
     ) -> List[Dict[str, Any]]:
@@ -133,7 +306,7 @@ class BaseSTTService(BaseService):
         pass
     
     @abstractmethod
-    async def detect_language(self, audio_file: Union[str, BinaryIO]) -> Dict[str, Any]:
+    async def detect_language(self, audio_file: Union[str, BinaryIO, bytes]) -> Dict[str, Any]:
         """
         Detect language of audio file
         

isa_model/inference/services/audio/openai_stt_service.py

@@ -47,7 +47,7 @@ class OpenAISTTService(BaseSTTService):
         wait=wait_exponential(multiplier=1, min=4, max=10),
         reraise=True
     )
-    async def transcribe(self, audio_file: Union[str, BinaryIO], language: Optional[str] = None, prompt: Optional[str] = None) -> Dict[str, Any]:
+    async def transcribe(self, audio_file: Union[str, BinaryIO, bytes], language: Optional[str] = None, prompt: Optional[str] = None, **kwargs) -> Dict[str, Any]:
         """
         Transcribe audio file to text using OpenAI's Whisper model.
         
@@ -73,8 +73,24 @@ class OpenAISTTService(BaseSTTService):
             if prompt:
                 transcription_params["prompt"] = prompt
             
-            # Handle file input - support base64 strings, file paths, and file objects
-            if isinstance(audio_file, str):
+            # Handle file input - support bytes, base64 strings, file paths, and file objects
+            if isinstance(audio_file, bytes):
+                # Handle bytes data directly
+                logger.info(f"Processing bytes audio data ({len(audio_file)} bytes)")
+                from io import BytesIO
+                audio_buffer = BytesIO(audio_file)
+                
+                # Use filename from kwargs if provided, otherwise default to .mp3
+                filename = kwargs.get('filename', 'audio.mp3')
+                if filename and not filename.endswith(('.mp3', '.wav', '.m4a', '.flac', '.ogg', '.webm', '.mp4')):
+                    filename += '.mp3'  # Add extension if missing
+                audio_buffer.name = filename
+                logger.info(f"Using filename: {filename}")
+                transcription = await self.client.audio.transcriptions.create(
+                    file=audio_buffer,
+                    **transcription_params
+                )
+            elif isinstance(audio_file, str):
                 # Check if it's a base64 string or file path
                 if len(audio_file) > 100 and not os.path.exists(audio_file):
                     # Likely a base64 string
@@ -147,7 +163,7 @@ class OpenAISTTService(BaseSTTService):
         wait=wait_exponential(multiplier=1, min=4, max=10),
         reraise=True
     )
-    async def translate(self, audio_file: Union[str, BinaryIO]) -> Dict[str, Any]:
+    async def translate(self, audio_file: Union[str, BinaryIO, bytes]) -> Dict[str, Any]:
         """
         Translate audio file to English text using OpenAI's Whisper model.
         
@@ -211,7 +227,7 @@ class OpenAISTTService(BaseSTTService):
             logger.error(f"Translation failed: {e}")
             raise
 
-    async def transcribe_batch(self, audio_files: List[Union[str, BinaryIO]], language: Optional[str] = None, prompt: Optional[str] = None) -> List[Dict[str, Any]]:
+    async def transcribe_batch(self, audio_files: List[Union[str, BinaryIO, bytes]], language: Optional[str] = None, prompt: Optional[str] = None) -> List[Dict[str, Any]]:
         """
         Transcribe multiple audio files in batch.
         
@@ -238,7 +254,7 @@ class OpenAISTTService(BaseSTTService):
         
         return results
 
-    async def detect_language(self, audio_file: Union[str, BinaryIO]) -> Dict[str, Any]:
+    async def detect_language(self, audio_file: Union[str, BinaryIO, bytes]) -> Dict[str, Any]:
         """
         Detect the language of an audio file.
         

isa_model/inference/services/custom_model_manager.py

@@ -0,0 +1,277 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+
+"""
+Custom Model Manager - Handles registration and management of custom trained models
+Provides integration for models trained through ISA Model training pipeline
+"""
+
+import logging
+import json
+import os
+from typing import Dict, List, Any, Optional
+from datetime import datetime
+from dataclasses import dataclass, asdict
+
+logger = logging.getLogger(__name__)
+
+@dataclass
+class CustomModelInfo:
+    """Information about a custom model"""
+    model_id: str
+    model_name: str
+    model_type: str  # 'text', 'vision', 'audio', etc.
+    provider: str
+    base_model: str  # The base model this was fine-tuned from
+    training_date: str
+    model_path: str  # Local path or HuggingFace repo
+    metadata: Dict[str, Any]
+    capabilities: List[str]
+    performance_metrics: Optional[Dict[str, float]] = None
+    deployment_config: Optional[Dict[str, Any]] = None
+    
+    def to_dict(self) -> Dict[str, Any]:
+        return asdict(self)
+
+class CustomModelManager:
+    """
+    Manages custom trained models in the ISA Model ecosystem
+    Handles registration, discovery, and integration of custom models
+    """
+    
+    def __init__(self, models_registry_path: str = None):
+        self.models_registry_path = models_registry_path or os.path.join(
+            os.path.expanduser("~"), ".isa_model", "custom_models.json"
+        )
+        self._models: Dict[str, CustomModelInfo] = {}
+        self._load_models_registry()
+    
+    def _load_models_registry(self):
+        """Load custom models registry from file"""
+        if os.path.exists(self.models_registry_path):
+            try:
+                with open(self.models_registry_path, 'r', encoding='utf-8') as f:
+                    models_data = json.load(f)
+                
+                for model_data in models_data.get('models', []):
+                    model_info = CustomModelInfo(**model_data)
+                    self._models[model_info.model_id] = model_info
+                
+                logger.info(f"Loaded {len(self._models)} custom models from registry")
+            except Exception as e:
+                logger.warning(f"Failed to load models registry: {e}")
+                self._models = {}
+        else:
+            # Create default registry with some ISA models
+            self._create_default_registry()
+    
+    def _create_default_registry(self):
+        """Create default registry with ISA models"""
+        default_models = [
+            CustomModelInfo(
+                model_id="isa-llm-service",
+                model_name="ISA LLM Service",
+                model_type="text",
+                provider="isa",
+                base_model="DialoGPT-small",
+                training_date="2024-12-19",
+                model_path="modal://isa-llm-inference",
+                metadata={
+                    "description": "ISA custom LLM service with fallback support",
+                    "parameters": "124M",
+                    "context_length": 1024,
+                    "languages": ["en", "zh"]
+                },
+                capabilities=["chat", "text_generation", "conversation"],
+                performance_metrics={
+                    "perplexity": 3.2,
+                    "bleu_score": 0.75,
+                    "response_time_ms": 850
+                },
+                deployment_config={
+                    "platform": "modal",
+                    "gpu_type": "A10G",
+                    "memory_gb": 16,
+                    "concurrent_requests": 5
+                }
+            ),
+            CustomModelInfo(
+                model_id="xenodennis/dialoGPT-small-20241219-v1",
+                model_name="ISA Fine-tuned DialoGPT",
+                model_type="text",
+                provider="huggingface",
+                base_model="microsoft/DialoGPT-small",
+                training_date="2024-12-19",
+                model_path="xenodennis/dialoGPT-small-20241219-v1",
+                metadata={
+                    "description": "DialoGPT model fine-tuned with ISA training pipeline",
+                    "parameters": "124M",
+                    "trainable_parameters": "294K (LoRA)",
+                    "training_steps": 1000,
+                    "languages": ["en", "zh"]
+                },
+                capabilities=["chat", "text_generation", "dialogue"],
+                performance_metrics={
+                    "final_loss": 2.1234,
+                    "eval_loss": 2.3456,
+                    "training_time_minutes": 15
+                }
+            ),
+            CustomModelInfo(
+                model_id="isa-custom-embeddings",
+                model_name="ISA Custom Embeddings",
+                model_type="embedding",
+                provider="isa",
+                base_model="sentence-transformers/all-MiniLM-L6-v2",
+                training_date="2024-12-19",
+                model_path="local://models/isa-embeddings",
+                metadata={
+                    "description": "Custom embeddings trained on ISA domain data",
+                    "dimensions": 384,
+                    "max_sequence_length": 512
+                },
+                capabilities=["embed", "similarity", "clustering"]
+            )
+        ]
+        
+        for model in default_models:
+            self._models[model.model_id] = model
+        
+        self._save_models_registry()
+        logger.info(f"Created default registry with {len(default_models)} models")
+    
+    def _save_models_registry(self):
+        """Save models registry to file"""
+        try:
+            os.makedirs(os.path.dirname(self.models_registry_path), exist_ok=True)
+            
+            registry_data = {
+                "version": "1.0",
+                "last_updated": datetime.now().isoformat(),
+                "models": [model.to_dict() for model in self._models.values()]
+            }
+            
+            with open(self.models_registry_path, 'w', encoding='utf-8') as f:
+                json.dump(registry_data, f, indent=2, ensure_ascii=False)
+            
+            logger.debug(f"Saved models registry to {self.models_registry_path}")
+        except Exception as e:
+            logger.error(f"Failed to save models registry: {e}")
+    
+    def register_model(self, model_info: CustomModelInfo) -> bool:
+        """Register a new custom model"""
+        try:
+            self._models[model_info.model_id] = model_info
+            self._save_models_registry()
+            logger.info(f"Registered custom model: {model_info.model_id}")
+            return True
+        except Exception as e:
+            logger.error(f"Failed to register model {model_info.model_id}: {e}")
+            return False
+    
+    def unregister_model(self, model_id: str) -> bool:
+        """Unregister a custom model"""
+        if model_id in self._models:
+            del self._models[model_id]
+            self._save_models_registry()
+            logger.info(f"Unregistered custom model: {model_id}")
+            return True
+        return False
+    
+    def get_model(self, model_id: str) -> Optional[CustomModelInfo]:
+        """Get custom model information"""
+        return self._models.get(model_id)
+    
+    def list_models(self, model_type: str = None, provider: str = None) -> List[CustomModelInfo]:
+        """List custom models with optional filtering"""
+        models = list(self._models.values())
+        
+        if model_type:
+            models = [m for m in models if m.model_type == model_type]
+        
+        if provider:
+            models = [m for m in models if m.provider == provider]
+        
+        return models
+    
+    def get_models_for_api(self) -> List[Dict[str, Any]]:
+        """Get models in API format for model listing"""
+        api_models = []
+        
+        for model in self._models.values():
+            api_model = {
+                "model_id": model.model_id,
+                "service_type": model.model_type,
+                "provider": model.provider,
+                "description": model.metadata.get("description", ""),
+                "capabilities": model.capabilities,
+                "custom": True,
+                "base_model": model.base_model,
+                "training_date": model.training_date
+            }
+            
+            # Add performance metrics if available
+            if model.performance_metrics:
+                api_model["performance"] = model.performance_metrics
+            
+            api_models.append(api_model)
+        
+        return api_models
+    
+    def search_models(self, query: str) -> List[CustomModelInfo]:
+        """Search custom models by query"""
+        query_lower = query.lower()
+        matching_models = []
+        
+        for model in self._models.values():
+            # Search in model_id, name, description, and capabilities
+            searchable_text = f"{model.model_id} {model.model_name} {model.metadata.get('description', '')} {' '.join(model.capabilities)}".lower()
+            
+            if query_lower in searchable_text:
+                matching_models.append(model)
+        
+        return matching_models
+    
+    def get_deployment_config(self, model_id: str) -> Optional[Dict[str, Any]]:
+        """Get deployment configuration for a model"""
+        model = self.get_model(model_id)
+        return model.deployment_config if model else None
+    
+    def update_performance_metrics(self, model_id: str, metrics: Dict[str, float]) -> bool:
+        """Update performance metrics for a model"""
+        model = self.get_model(model_id)
+        if model:
+            model.performance_metrics = metrics
+            self._save_models_registry()
+            return True
+        return False
+    
+    def get_provider_models(self, provider: str) -> List[CustomModelInfo]:
+        """Get all models for a specific provider"""
+        return [model for model in self._models.values() if model.provider == provider]
+    
+    def get_stats(self) -> Dict[str, Any]:
+        """Get statistics about custom models"""
+        models_by_type = {}
+        models_by_provider = {}
+        
+        for model in self._models.values():
+            models_by_type[model.model_type] = models_by_type.get(model.model_type, 0) + 1
+            models_by_provider[model.provider] = models_by_provider.get(model.provider, 0) + 1
+        
+        return {
+            "total_models": len(self._models),
+            "models_by_type": models_by_type,
+            "models_by_provider": models_by_provider,
+            "registry_path": self.models_registry_path
+        }
+
+# Global instance
+_custom_model_manager = None
+
+def get_custom_model_manager() -> CustomModelManager:
+    """Get the global custom model manager instance"""
+    global _custom_model_manager
+    if _custom_model_manager is None:
+        _custom_model_manager = CustomModelManager()
+    return _custom_model_manager

isa_model/inference/services/embedding/ollama_embed_service.py

@@ -4,6 +4,7 @@ import asyncio
 from typing import List, Dict, Any, Optional
 
 from isa_model.inference.services.embedding.base_embed_service import BaseEmbedService
+from isa_model.core.config.config_manager import ConfigManager
 
 logger = logging.getLogger(__name__)
 
@@ -21,9 +22,20 @@ class OllamaEmbedService(BaseEmbedService):
         
         # Initialize HTTP client with provider configuration
         try:
-            host = provider_config.get("host", "localhost")
-            port = provider_config.get("port", 11434)
-            base_url = f"http://{host}:{port}"
+            config_manager = ConfigManager()
+            # Use Consul discovery with fallback
+            default_base_url = config_manager.get_ollama_url()
+            
+            if "base_url" in provider_config:
+                base_url = provider_config["base_url"]
+            else:
+                host = provider_config.get("host", "localhost")
+                port = provider_config.get("port", 11434)
+                base_url = provider_config.get("base_url", f"http://{host}:{port}")
+                
+            # Use config manager default (Consul discovery) if still not set
+            if base_url == f"http://localhost:11434":
+                base_url = default_base_url
             
             self.client = httpx.AsyncClient(base_url=base_url, timeout=30.0)