Update app/config.py

app/config.py

@@ -1,23 +1,21 @@
+"""
+Configuration settings for Universal Model Trainer
+"""
+
 from pydantic_settings import BaseSettings
-from typing import List, Optional, Dict, Any
+from typing import Optional, List
+from functools import lru_cache
 import os
 
 
 class Settings(BaseSettings):
+    """Application settings loaded from environment variables."""
+    
     # Application
     APP_NAME: str = "Universal Model Trainer"
-    APP_VERSION: str = "1.0.0"
-    DEBUG: bool = os.getenv("DEBUG", "false").lower() == "true"
-    
-    # Authentication
-    APP_PASSWORD: str = os.getenv("APP_PASSWORD", "")  # Set via HF Space secret
-    SESSION_SECRET_KEY: str = os.getenv("SESSION_SECRET_KEY", "change-this-in-production-use-32-chars")
-    SESSION_EXPIRE_HOURS: int = 24
-    
-    @property
-    def is_auth_configured(self) -> bool:
-        """Check if password authentication is configured"""
-        return bool(self.APP_PASSWORD)
+    VERSION: str = "1.0.0"
+    DEBUG: bool = False
+    SECRET_KEY: str = "change-me-in-production"
     
     # Server
     HOST: str = "0.0.0.0"
@@ -25,63 +23,60 @@ class Settings(BaseSettings):
     WORKERS: int = 1
     
     # HuggingFace
-    HF_TOKEN: Optional[str] = os.getenv("HF_TOKEN")
-    HF_HOME: str = os.getenv("HF_HOME", "/app/.cache/huggingface")
+    HF_TOKEN: Optional[str] = None
+    HF_USERNAME: Optional[str] = None
+    HF_CACHE_DIR: str = "/app/cache"
     
-    # Redis/Queue
-    REDIS_URL: str = os.getenv("REDIS_URL", "redis://localhost:6379/0")
-    MAX_CONCURRENT_JOBS: int = 1
-    JOB_TIMEOUT_HOURS: int = 24
+    # Redis / Queue
+    REDIS_URL: str = "redis://localhost:6379/0"
+    QUEUE_NAME: str = "training_queue"
+    MAX_CONCURRENT_JOBS: int = 3
     
     # Database
-    DATABASE_URL: str = os.getenv("DATABASE_URL", "sqlite+aiosqlite:///./data/trainer.db")
+    DATABASE_URL: str = "sqlite:///./data/trainer.db"
     
     # Storage Paths
+    UPLOAD_DIR: str = "/app/uploads"
     MODELS_DIR: str = "/app/models"
-    DATASETS_DIR: str = "/app/datasets"
-    OUTPUT_DIR: str = "/app/outputs"
-    CACHE_DIR: str = "/app/.cache"
     LOGS_DIR: str = "/app/logs"
+    CACHE_DIR: str = "/app/cache"
     
     # Training Defaults
-    DEFAULT_EPOCHS: int = 3
-    DEFAULT_BATCH_SIZE: int = 1
+    DEFAULT_BATCH_SIZE: int = 8
     DEFAULT_LEARNING_RATE: float = 5e-5
     DEFAULT_MAX_LENGTH: int = 512
-    DEFAULT_WARMUP_STEPS: int = 100
-    DEFAULT_WEIGHT_DECAY: float = 0.01
+    DEFAULT_EPOCHS: int = 3
+    DEFAULT_WARMUP_RATIO: float = 0.1
     
     # Hardware
-    DEVICE: str = "cpu"
-    MIXED_PRECISION: bool = False
-    GRADIENT_CHECKPOINTING: bool = True
-    
-    # PEFT/LoRA Defaults
-    USE_PEFT_BY_DEFAULT: bool = True
-    DEFAULT_LORA_R: int = 16
-    DEFAULT_LORA_ALPHA: int = 32
-    DEFAULT_LORA_DROPOUT: float = 0.05
-    DEFAULT_TARGET_MODULES: List[str] = ["q_proj", "v_proj"]
+    DEVICE: str = "auto"  # auto, cpu, cuda, mps
+    DTYPE: str = "auto"   # auto, float32, float16, bfloat16
+    
+    # PEFT / LoRA Defaults
+    PEFT_DEFAULTS: dict = {
+        "lora_r": 16,
+        "lora_alpha": 32,
+        "lora_dropout": 0.05,
+        "target_modules": ["q_proj", "v_proj"]
+    }
     
     # Model Support
     SUPPORTED_TASKS: List[str] = [
         "causal-lm",
-        "seq2seq", 
+        "seq2seq",
         "token-classification",
-        "text-classification",
+        "sequence-classification",
         "question-answering",
-        "reasoning"
+        "summarization",
+        "translation",
+        "text-classification",
+        "masked-lm",
+        "vision-classification",
+        "vision-segmentation",
+        "audio-classification",
+        "audio-transcription"
     ]
     
-    DEFAULT_MODELS: Dict[str, str] = {
-        "causal-lm": "gpt2",
-        "seq2seq": "t5-small",
-        "token-classification": "bert-base-uncased",
-        "text-classification": "distilbert-base-uncased",
-        "question-answering": "distilbert-base-uncased-distilled-squad",
-        "reasoning": "gpt2"
-    }
-    
     # Dataset Sources
     DATASET_SOURCES: List[str] = [
         "huggingface",
@@ -91,89 +86,159 @@ class Settings(BaseSettings):
     ]
     
     # Logging
-    LOG_LEVEL: str = os.getenv("LOG_LEVEL", "INFO")
-    LOG_FORMAT: str = "<green>{time:YYYY-MM-DD HH:mm:ss}</green> | <level>{level: <8}</level> | <cyan>{name}</cyan>:<cyan>{function}</cyan>:<cyan>{line}</cyan> - <level>{message}</level>"
+    LOG_LEVEL: str = "INFO"
+    LOG_FORMAT: str = "%(asctime)s - %(name)s - %(levelname)s - %(message)s"
     
     # Experiment Tracking
-    WANDB_PROJECT: Optional[str] = os.getenv("WANDB_PROJECT")
-    WANDB_API_KEY: Optional[str] = os.getenv("WANDB_API_KEY")
-    MLFLOW_TRACKING_URI: Optional[str] = os.getenv("MLFLOW_TRACKING_URI")
+    WANDB_API_KEY: Optional[str] = None
+    WANDB_PROJECT: str = "universal-model-trainer"
     
     # Security
-    CORS_ORIGINS: List[str] = ["*"]
     API_KEY_HEADER: str = "X-API-Key"
+    ENABLE_AUTH: bool = False
     
     # Rate Limiting
-    RATE_LIMIT_REQUESTS: int = 100
-    RATE_LIMIT_PERIOD_SECONDS: int = 60
+    RATE_LIMIT_PER_MINUTE: int = 60
     
     # Resource Limits
-    MAX_DATASET_SIZE_GB: float = 10.0
-    MAX_MODEL_SIZE_GB: float = 5.0
-    MAX_TRAINING_TIME_HOURS: int = 24
+    MAX_UPLOAD_SIZE_MB: int = 500
+    MAX_DATASET_SIZE_MB: int = 10000
+    MAX_TRAINING_HOURS: int = 24
     
     class Config:
         env_file = ".env"
+        env_file_encoding = "utf-8"
         case_sensitive = True
 
 
-# Training templates for different tasks
-TRAINING_TEMPLATES: Dict[str, Dict[str, Any]] = {
+@lru_cache()
+def get_settings() -> Settings:
+    """Get cached settings instance."""
+    return Settings()
+
+
+# Export settings instance
+settings = get_settings()
+
+
+# Training Configuration Templates
+TRAINING_TEMPLATES = {
     "causal-lm": {
-        "name": "Causal Language Model",
-        "description": "Text generation with GPT-style models",
+        "task": "causal-lm",
+        "model_type": "decoder",
         "default_model": "gpt2",
         "default_args": {
-            "max_length": 512,
-            "truncate": True,
-            "padding": "max_length"
+            "per_device_train_batch_size": 4,
+            "gradient_accumulation_steps": 4,
+            "learning_rate": 5e-5,
+            "max_grad_norm": 1.0,
+            "warmup_ratio": 0.1,
+            "weight_decay": 0.01
         }
     },
     "seq2seq": {
-        "name": "Sequence-to-Sequence",
-        "description": "Summarization, translation, etc.",
+        "task": "seq2seq",
+        "model_type": "encoder-decoder",
         "default_model": "t5-small",
         "default_args": {
-            "max_source_length": 512,
-            "max_target_length": 128
+            "per_device_train_batch_size": 8,
+            "learning_rate": 5e-5,
+            "warmup_ratio": 0.1,
+            "source_lang": "en",
+            "target_lang": "en"
         }
     },
     "token-classification": {
-        "name": "Token Classification",
-        "description": "Named Entity Recognition, POS Tagging",
+        "task": "token-classification",
+        "model_type": "encoder",
         "default_model": "bert-base-uncased",
         "default_args": {
-            "max_length": 512,
-            "label_all_tokens": False
+            "per_device_train_batch_size": 16,
+            "learning_rate": 3e-5,
+            "label_column": "ner_tags"
         }
     },
-    "text-classification": {
-        "name": "Text Classification",
-        "description": "Sentiment analysis, categorization",
+    "sequence-classification": {
+        "task": "sequence-classification",
+        "model_type": "encoder",
         "default_model": "distilbert-base-uncased",
         "default_args": {
-            "max_length": 512
+            "per_device_train_batch_size": 16,
+            "learning_rate": 2e-5,
+            "num_labels": 2,
+            "label_column": "label"
         }
     },
     "question-answering": {
-        "name": "Question Answering",
-        "description": "Extractive question answering",
+        "task": "question-answering",
+        "model_type": "encoder",
         "default_model": "distilbert-base-uncased-distilled-squad",
         "default_args": {
-            "max_length": 384,
-            "doc_stride": 128
+            "per_device_train_batch_size": 12,
+            "learning_rate": 3e-5,
+            "max_answer_length": 30
         }
     },
-    "reasoning": {
-        "name": "Reasoning/Chain-of-Thought",
-        "description": "Step-by-step reasoning models",
-        "default_model": "gpt2",
+    "text-classification": {
+        "task": "text-classification",
+        "model_type": "encoder",
+        "default_model": "roberta-base",
         "default_args": {
-            "max_length": 1024,
-            "include_reasoning": True
+            "per_device_train_batch_size": 16,
+            "learning_rate": 1e-5,
+            "num_labels": 2
+        }
+    },
+    "summarization": {
+        "task": "summarization",
+        "model_type": "encoder-decoder",
+        "default_model": "facebook/bart-large-cnn",
+        "default_args": {
+            "per_device_train_batch_size": 4,
+            "learning_rate": 3e-5,
+            "max_source_length": 1024,
+            "max_target_length": 128
+        }
+    },
+    "translation": {
+        "task": "translation",
+        "model_type": "encoder-decoder",
+        "default_model": "Helsinki-NLP/opus-mt-en-fr",
+        "default_args": {
+            "per_device_train_batch_size": 8,
+            "learning_rate": 5e-5,
+            "source_lang": "en",
+            "target_lang": "fr"
+        }
+    },
+    "masked-lm": {
+        "task": "masked-lm",
+        "model_type": "encoder",
+        "default_model": "bert-base-uncased",
+        "default_args": {
+            "per_device_train_batch_size": 16,
+            "learning_rate": 5e-5,
+            "mlm_probability": 0.15
+        }
+    },
+    "vision-classification": {
+        "task": "vision-classification",
+        "model_type": "vision",
+        "default_model": "google/vit-base-patch16-224",
+        "default_args": {
+            "per_device_train_batch_size": 32,
+            "learning_rate": 2e-5,
+            "image_column": "image"
+        }
+    },
+    "audio-transcription": {
+        "task": "audio-transcription",
+        "model_type": "audio",
+        "default_model": "openai/whisper-small",
+        "default_args": {
+            "per_device_train_batch_size": 8,
+            "learning_rate": 1e-5,
+            "audio_column": "audio"
         }
     }
-}
-
-
-settings = Settings()
+}