langvision 0.0.1__py3-none-any.whl → 0.1.0__py3-none-any.whl

langvision/__init__.py

@@ -1,7 +1,82 @@
 """
-langvision - A platform for building and deploying AI agents
+langvision - Modular Vision LLMs with Efficient LoRA Fine-Tuning
+
+A research-friendly framework for building and fine-tuning Vision Large Language Models
+with efficient Low-Rank Adaptation (LoRA) support.
 """
 
 __version__ = "0.1.0"
+__author__ = "Pritesh Raj"
+__email__ = "priteshraj10@gmail.com"
+
+# Core imports for easy access
+from .models.vision_transformer import VisionTransformer
+from .models.lora import LoRALinear, LoRAConfig, AdaLoRALinear, QLoRALinear
+from .models.resnet import resnet18, resnet34, resnet50, resnet101, resnet152
+from .models.multimodal import VisionLanguageModel, create_multimodal_model, CLIPLoss
+from .utils.config import default_config
+from .training.trainer import Trainer
+from .training.advanced_trainer import AdvancedTrainer, TrainingConfig
+from .data.datasets import get_dataset
+from .data.enhanced_datasets import (
+    EnhancedImageDataset, MultimodalDataset, DatasetConfig, 
+    create_enhanced_dataloaders, SmartAugmentation
+)
+from .utils.metrics import (
+    MetricsTracker, ClassificationMetrics, ContrastiveMetrics, 
+    EvaluationSuite, PerformanceMetrics
+)
+from .callbacks.base import Callback, CallbackManager
+from .concepts import RLHF, CoT, CCoT, GRPO, RLVR, DPO, PPO, LIME, SHAP
+
+# Version info
+__all__ = [
+    "__version__",
+    "__author__",
+    "__email__",
+    # Core Models
+    "VisionTransformer",
+    "resnet18", "resnet34", "resnet50", "resnet101", "resnet152",
+    "VisionLanguageModel", "create_multimodal_model",
+    # LoRA Components
+    "LoRALinear", "LoRAConfig", "AdaLoRALinear", "QLoRALinear",
+    # Training
+    "Trainer", "AdvancedTrainer", "TrainingConfig",
+    # Data
+    "get_dataset", "EnhancedImageDataset", "MultimodalDataset", 
+    "DatasetConfig", "create_enhanced_dataloaders", "SmartAugmentation",
+    # Utilities
+    "default_config", "MetricsTracker", "ClassificationMetrics", 
+    "ContrastiveMetrics", "EvaluationSuite", "PerformanceMetrics",
+    # Callbacks
+    "Callback", "CallbackManager",
+    # Loss Functions
+    "CLIPLoss",
+    # Concepts
+    "RLHF", "CoT", "CCoT", "GRPO", "RLVR", "DPO", "PPO", "LIME", "SHAP",
+]
+
+# Optional imports for advanced usage
+try:
+    from .callbacks import EarlyStoppingCallback, LoggingCallback
+    from .utils.device import get_device, to_device
+    __all__.extend([
+        "EarlyStoppingCallback",
+        "LoggingCallback", 
+        "get_device",
+        "to_device"
+    ])
+except ImportError:
+    pass
 
-"""langvision: Vision LLMs with LoRA and unified model zoo.""" 
+# Package metadata
+PACKAGE_METADATA = {
+    "name": "langvision",
+    "version": __version__,
+    "description": "Modular Vision LLMs with Efficient LoRA Fine-Tuning",
+    "author": __author__,
+    "email": __email__,
+    "url": "https://github.com/langtrain-ai/langtrain",
+    "license": "MIT",
+    "python_requires": ">=3.8",
+} 

langvision/callbacks/base.py

@@ -1,11 +1,170 @@
-class Callback:
-    def on_train_begin(self, trainer):
+"""Enhanced callback system for training and evaluation hooks."""
+
+from abc import ABC, abstractmethod
+from typing import Dict, Any, Optional, List
+import logging
+import traceback
+
+
+class Callback(ABC):
+    """Base class for all callbacks with comprehensive training hooks."""
+    
+    def __init__(self, name: Optional[str] = None):
+        self.name = name or self.__class__.__name__
+        self.logger = logging.getLogger(f"langvision.callbacks.{self.name}")
+    
+    def on_train_start(self, trainer) -> None:
+        """Called at the beginning of training."""
         pass
-    def on_epoch_begin(self, trainer, epoch):
+    
+    def on_train_end(self, trainer) -> None:
+        """Called at the end of training."""
         pass
-    def on_batch_end(self, trainer, batch, logs=None):
+    
+    def on_epoch_start(self, trainer, epoch: int) -> None:
+        """Called at the beginning of each epoch."""
         pass
-    def on_epoch_end(self, trainer, epoch, logs=None):
+    
+    def on_epoch_end(self, trainer, epoch: int, metrics: Dict[str, Any]) -> None:
+        """Called at the end of each epoch."""
         pass
-    def on_train_end(self, trainer):
-        pass 
+    
+    def on_batch_start(self, trainer, batch_idx: int, batch: Dict[str, Any]) -> None:
+        """Called at the beginning of each batch."""
+        pass
+    
+    def on_batch_end(self, trainer, batch_idx: int, batch: Dict[str, Any], outputs: Dict[str, Any]) -> None:
+        """Called at the end of each batch."""
+        pass
+    
+    def on_validation_start(self, trainer) -> None:
+        """Called at the beginning of validation."""
+        pass
+    
+    def on_validation_end(self, trainer, metrics: Dict[str, Any]) -> None:
+        """Called at the end of validation."""
+        pass
+    
+    def on_test_start(self, trainer) -> None:
+        """Called at the beginning of testing."""
+        pass
+    
+    def on_test_end(self, trainer, metrics: Dict[str, Any]) -> None:
+        """Called at the end of testing."""
+        pass
+    
+    def on_checkpoint_save(self, trainer, checkpoint_path: str, metrics: Dict[str, Any]) -> None:
+        """Called when a checkpoint is saved."""
+        pass
+    
+    def on_checkpoint_load(self, trainer, checkpoint_path: str) -> None:
+        """Called when a checkpoint is loaded."""
+        pass
+    
+    def on_lr_schedule(self, trainer, old_lr: float, new_lr: float) -> None:
+        """Called when learning rate is scheduled."""
+        pass
+    
+    def on_exception(self, trainer, exception: Exception) -> bool:
+        """Called when an exception occurs during training.
+        
+        Returns:
+            bool: True if the exception was handled and training should continue,
+                  False if training should stop.
+        """
+        return False
+
+
+class CallbackManager:
+    """Manager for handling multiple callbacks with error handling."""
+    
+    def __init__(self, callbacks: Optional[List[Callback]] = None):
+        self.callbacks = callbacks or []
+        self.logger = logging.getLogger("langvision.callbacks.manager")
+    
+    def add_callback(self, callback: Callback) -> None:
+        """Add a callback to the manager."""
+        if not isinstance(callback, Callback):
+            raise TypeError(f"Expected Callback instance, got {type(callback)}")
+        self.callbacks.append(callback)
+        self.logger.info(f"Added callback: {callback.name}")
+    
+    def remove_callback(self, callback_name: str) -> bool:
+        """Remove a callback by name."""
+        for i, callback in enumerate(self.callbacks):
+            if callback.name == callback_name:
+                del self.callbacks[i]
+                self.logger.info(f"Removed callback: {callback_name}")
+                return True
+        return False
+    
+    def _call_callbacks(self, method_name: str, *args, **kwargs) -> None:
+        """Safely call a method on all callbacks."""
+        for callback in self.callbacks:
+            try:
+                method = getattr(callback, method_name, None)
+                if method and callable(method):
+                    method(*args, **kwargs)
+            except Exception as e:
+                self.logger.error(
+                    f"Error in callback {callback.name}.{method_name}: {str(e)}\n"
+                    f"Traceback: {traceback.format_exc()}"
+                )
+                # Continue with other callbacks even if one fails
+    
+    def on_train_start(self, trainer) -> None:
+        self._call_callbacks('on_train_start', trainer)
+    
+    def on_train_end(self, trainer) -> None:
+        self._call_callbacks('on_train_end', trainer)
+    
+    def on_epoch_start(self, trainer, epoch: int) -> None:
+        self._call_callbacks('on_epoch_start', trainer, epoch)
+    
+    def on_epoch_end(self, trainer, epoch: int, metrics: Dict[str, Any]) -> None:
+        self._call_callbacks('on_epoch_end', trainer, epoch, metrics)
+    
+    def on_batch_start(self, trainer, batch_idx: int, batch: Dict[str, Any]) -> None:
+        self._call_callbacks('on_batch_start', trainer, batch_idx, batch)
+    
+    def on_batch_end(self, trainer, batch_idx: int, batch: Dict[str, Any], outputs: Dict[str, Any]) -> None:
+        self._call_callbacks('on_batch_end', trainer, batch_idx, batch, outputs)
+    
+    def on_validation_start(self, trainer) -> None:
+        self._call_callbacks('on_validation_start', trainer)
+    
+    def on_validation_end(self, trainer, metrics: Dict[str, Any]) -> None:
+        self._call_callbacks('on_validation_end', trainer, metrics)
+    
+    def on_test_start(self, trainer) -> None:
+        self._call_callbacks('on_test_start', trainer)
+    
+    def on_test_end(self, trainer, metrics: Dict[str, Any]) -> None:
+        self._call_callbacks('on_test_end', trainer, metrics)
+    
+    def on_checkpoint_save(self, trainer, checkpoint_path: str, metrics: Dict[str, Any]) -> None:
+        self._call_callbacks('on_checkpoint_save', trainer, checkpoint_path, metrics)
+    
+    def on_checkpoint_load(self, trainer, checkpoint_path: str) -> None:
+        self._call_callbacks('on_checkpoint_load', trainer, checkpoint_path)
+    
+    def on_lr_schedule(self, trainer, old_lr: float, new_lr: float) -> None:
+        self._call_callbacks('on_lr_schedule', trainer, old_lr, new_lr)
+    
+    def on_exception(self, trainer, exception: Exception) -> bool:
+        """Handle exceptions through callbacks.
+        
+        Returns:
+            bool: True if any callback handled the exception and training should continue.
+        """
+        handled = False
+        for callback in self.callbacks:
+            try:
+                if callback.on_exception(trainer, exception):
+                    handled = True
+                    self.logger.info(f"Exception handled by callback: {callback.name}")
+            except Exception as callback_error:
+                self.logger.error(
+                    f"Error in callback {callback.name}.on_exception: {str(callback_error)}"
+                )
+        return handled 

langvision/cli/__init__.py

@@ -0,0 +1,85 @@
+import argparse
+import sys
+from .train import main as train_main
+from .finetune import main as finetune_main
+from .evaluate import main as evaluate_main
+from .export import main as export_main
+from .model_zoo import main as model_zoo_main
+from .config import main as config_main
+
+__version__ = "0.1.0"  # Keep in sync with package version in pyproject.toml
+
+def print_banner():
+    banner = r"""
+\033[1;36m
+ _                     __     ___     _             
+| |    __ _ _ __   __ _\ \   / (_)___(_) ___  _ __  
+| |   / _` | '_ \ / _` |\ \ / /| / __| |/ _ \| '_ \ 
+| |__| (_| | | | | (_| | \ V / | \__ \ | (_) | | | |
+|_____\__,_|_| |_|\__, |  \_/  |_|___/_|\___/|_| |_|
+                  |___/                             
+\033[0m
+"""
+    print(banner)
+    print("\033[1;33mLANGVISION\033[0m: Modular Vision LLMs with Efficient LoRA Fine-Tuning")
+    print(f"\033[1;35mVersion:\033[0m {__version__}")
+    print("\033[1;32mDocs:\033[0m https://github.com/langtrain-ai/langtrain/tree/main/docs    \033[1;34mPyPI:\033[0m https://pypi.org/project/langvision/\n")
+
+def main():
+    print_banner()
+    parser = argparse.ArgumentParser(
+        prog="langvision",
+        description="Langvision: Modular Vision LLMs with Efficient LoRA Fine-Tuning.\n\nUse subcommands to train or finetune vision models."
+    )
+    parser.add_argument('--version', action='version', version=f'%(prog)s {__version__}')
+    subparsers = parser.add_subparsers(dest='command', required=True, help='Sub-commands')
+
+    # Train subcommand
+    train_parser = subparsers.add_parser('train', help='Train a VisionTransformer model')
+    train_parser.add_argument('args', nargs=argparse.REMAINDER)
+
+    # Finetune subcommand
+    finetune_parser = subparsers.add_parser('finetune', help='Finetune a VisionTransformer model with LoRA and LLM concepts')
+    finetune_parser.add_argument('args', nargs=argparse.REMAINDER)
+
+    # Evaluate subcommand
+    evaluate_parser = subparsers.add_parser('evaluate', help='Evaluate a trained model')
+    evaluate_parser.add_argument('args', nargs=argparse.REMAINDER)
+
+    # Export subcommand
+    export_parser = subparsers.add_parser('export', help='Export a model to various formats (ONNX, TorchScript)')
+    export_parser.add_argument('args', nargs=argparse.REMAINDER)
+
+    # Model Zoo subcommand
+    model_zoo_parser = subparsers.add_parser('model-zoo', help='Browse and download pre-trained models')
+    model_zoo_parser.add_argument('args', nargs=argparse.REMAINDER)
+
+    # Config subcommand
+    config_parser = subparsers.add_parser('config', help='Manage configuration files')
+    config_parser.add_argument('args', nargs=argparse.REMAINDER)
+
+    args = parser.parse_args()
+
+    if args.command == 'train':
+        sys.argv = [sys.argv[0]] + args.args
+        train_main()
+    elif args.command == 'finetune':
+        sys.argv = [sys.argv[0]] + args.args
+        finetune_main()
+    elif args.command == 'evaluate':
+        sys.argv = [sys.argv[0]] + args.args
+        evaluate_main()
+    elif args.command == 'export':
+        sys.argv = [sys.argv[0]] + args.args
+        export_main()
+    elif args.command == 'model-zoo':
+        sys.argv = [sys.argv[0]] + args.args
+        model_zoo_main()
+    elif args.command == 'config':
+        sys.argv = [sys.argv[0]] + args.args
+        config_main()
+    else:
+        parser.print_help()
+
+if __name__ == '__main__':
+    main()

langvision/cli/complete_cli.py

@@ -0,0 +1,319 @@
+"""
+Complete CLI interface for Langvision framework.
+"""
+
+import argparse
+import sys
+import logging
+from pathlib import Path
+from typing import Optional, Dict, Any
+
+from ..utils.setup import initialize_langvision, quick_setup
+from ..models.lora import LoRAConfig
+from ..training.advanced_trainer import AdvancedTrainer, TrainingConfig
+from ..data.enhanced_datasets import create_enhanced_dataloaders, DatasetConfig
+from ..models.vision_transformer import VisionTransformer
+from ..models.resnet import resnet50
+from ..models.multimodal import create_multimodal_model
+from ..utils.device import get_device
+
+
+def create_parser() -> argparse.ArgumentParser:
+    """Create comprehensive argument parser for Langvision CLI."""
+    
+    parser = argparse.ArgumentParser(
+        prog="langvision",
+        description="Langvision: Advanced Vision-Language Models with Efficient LoRA Fine-Tuning",
+        formatter_class=argparse.RawDescriptionHelpFormatter,
+        epilog="""
+Examples:
+  # Basic image classification training
+  langvision train --model vit --dataset cifar10 --data-dir ./data --epochs 50
+  
+  # LoRA fine-tuning with custom parameters
+  langvision train --model resnet50 --lora-r 16 --lora-alpha 32 --freeze-backbone
+  
+  # Multimodal training
+  langvision train-multimodal --vision-model vit --text-model bert-base-uncased --data-dir ./data
+  
+  # Evaluation
+  langvision evaluate --model-path ./checkpoints/best_model.pt --data-dir ./test_data
+        """
+    )
+    
+    parser.add_argument("--version", action="version", version="langvision 0.1.0")
+    
+    # Create subparsers
+    subparsers = parser.add_subparsers(dest="command", help="Available commands")
+    
+    # Training command
+    train_parser = subparsers.add_parser("train", help="Train a vision model")
+    add_training_args(train_parser)
+    
+    # Multimodal training command
+    multimodal_parser = subparsers.add_parser("train-multimodal", help="Train a multimodal vision-language model")
+    add_multimodal_args(multimodal_parser)
+    
+    # Evaluation command
+    eval_parser = subparsers.add_parser("evaluate", help="Evaluate a trained model")
+    add_evaluation_args(eval_parser)
+    
+    # Setup command
+    setup_parser = subparsers.add_parser("setup", help="Setup and validate Langvision environment")
+    add_setup_args(setup_parser)
+    
+    return parser
+
+
+def add_training_args(parser: argparse.ArgumentParser) -> None:
+    """Add training-specific arguments."""
+    
+    # Model arguments
+    model_group = parser.add_argument_group("Model Configuration")
+    model_group.add_argument("--model", choices=["vit", "resnet18", "resnet34", "resnet50", "resnet101", "resnet152"],
+                           default="vit", help="Model architecture to use")
+    model_group.add_argument("--num-classes", type=int, default=10, help="Number of output classes")
+    model_group.add_argument("--img-size", type=int, default=224, help="Input image size")
+    
+    # LoRA arguments
+    lora_group = parser.add_argument_group("LoRA Configuration")
+    lora_group.add_argument("--use-lora", action="store_true", help="Enable LoRA fine-tuning")
+    lora_group.add_argument("--lora-r", type=int, default=16, help="LoRA rank")
+    lora_group.add_argument("--lora-alpha", type=float, default=32.0, help="LoRA alpha")
+    lora_group.add_argument("--lora-dropout", type=float, default=0.1, help="LoRA dropout")
+    lora_group.add_argument("--freeze-backbone", action="store_true", help="Freeze backbone parameters")
+    
+    # Data arguments
+    data_group = parser.add_argument_group("Data Configuration")
+    data_group.add_argument("--dataset", choices=["cifar10", "cifar100", "imagefolder"], 
+                          default="cifar10", help="Dataset to use")
+    data_group.add_argument("--data-dir", type=str, required=True, help="Path to dataset")
+    data_group.add_argument("--batch-size", type=int, default=32, help="Batch size")
+    data_group.add_argument("--num-workers", type=int, default=4, help="Number of data loading workers")
+    data_group.add_argument("--augmentation", action="store_true", help="Enable data augmentation")
+    data_group.add_argument("--augmentation-strength", type=float, default=0.5, help="Augmentation strength")
+    
+    # Training arguments
+    train_group = parser.add_argument_group("Training Configuration")
+    train_group.add_argument("--epochs", type=int, default=100, help="Number of training epochs")
+    train_group.add_argument("--lr", type=float, default=1e-4, help="Learning rate")
+    train_group.add_argument("--weight-decay", type=float, default=1e-4, help="Weight decay")
+    train_group.add_argument("--optimizer", choices=["adam", "adamw", "sgd"], default="adamw", help="Optimizer")
+    train_group.add_argument("--scheduler", choices=["cosine", "step", "plateau"], default="cosine", help="LR scheduler")
+    train_group.add_argument("--warmup-epochs", type=int, default=5, help="Warmup epochs")
+    train_group.add_argument("--use-amp", action="store_true", help="Use mixed precision training")
+    train_group.add_argument("--gradient-clip", type=float, default=1.0, help="Gradient clipping norm")
+    
+    # Output arguments
+    output_group = parser.add_argument_group("Output Configuration")
+    output_group.add_argument("--output-dir", type=str, default="./outputs", help="Output directory")
+    output_group.add_argument("--experiment-name", type=str, default="langvision_experiment", help="Experiment name")
+    output_group.add_argument("--save-interval", type=int, default=5, help="Save checkpoint every N epochs")
+    output_group.add_argument("--log-interval", type=int, default=10, help="Log every N batches")
+    
+    # System arguments
+    system_group = parser.add_argument_group("System Configuration")
+    system_group.add_argument("--seed", type=int, default=42, help="Random seed")
+    system_group.add_argument("--device", type=str, default="auto", help="Device to use (auto, cpu, cuda, mps)")
+    system_group.add_argument("--log-level", choices=["DEBUG", "INFO", "WARNING", "ERROR"], 
+                            default="INFO", help="Logging level")
+
+
+def add_multimodal_args(parser: argparse.ArgumentParser) -> None:
+    """Add multimodal training-specific arguments."""
+    
+    # Inherit basic training args
+    add_training_args(parser)
+    
+    # Multimodal-specific arguments
+    multimodal_group = parser.add_argument_group("Multimodal Configuration")
+    multimodal_group.add_argument("--vision-model", choices=["vit_base"], default="vit_base", 
+                                 help="Vision model architecture")
+    multimodal_group.add_argument("--text-model", type=str, default="bert-base-uncased", 
+                                 help="Text model from HuggingFace")
+    multimodal_group.add_argument("--vision-dim", type=int, default=768, help="Vision feature dimension")
+    multimodal_group.add_argument("--text-dim", type=int, default=768, help="Text feature dimension")
+    multimodal_group.add_argument("--hidden-dim", type=int, default=512, help="Hidden dimension for fusion")
+    multimodal_group.add_argument("--max-text-length", type=int, default=77, help="Maximum text sequence length")
+    multimodal_group.add_argument("--annotations-file", type=str, help="Path to text annotations file")
+    multimodal_group.add_argument("--contrastive-weight", type=float, default=1.0, 
+                                 help="Weight for contrastive loss")
+    multimodal_group.add_argument("--classification-weight", type=float, default=0.5, 
+                                 help="Weight for classification loss")
+
+
+def add_evaluation_args(parser: argparse.ArgumentParser) -> None:
+    """Add evaluation-specific arguments."""
+    
+    parser.add_argument("--model-path", type=str, required=True, help="Path to trained model checkpoint")
+    parser.add_argument("--data-dir", type=str, required=True, help="Path to evaluation dataset")
+    parser.add_argument("--dataset", choices=["cifar10", "cifar100", "imagefolder"], 
+                       default="cifar10", help="Dataset type")
+    parser.add_argument("--batch-size", type=int, default=32, help="Batch size for evaluation")
+    parser.add_argument("--num-workers", type=int, default=4, help="Number of data loading workers")
+    parser.add_argument("--output-dir", type=str, default="./eval_results", help="Output directory for results")
+    parser.add_argument("--save-predictions", action="store_true", help="Save model predictions")
+    parser.add_argument("--benchmark", action="store_true", help="Run inference speed benchmark")
+    parser.add_argument("--device", type=str, default="auto", help="Device to use")
+
+
+def add_setup_args(parser: argparse.ArgumentParser) -> None:
+    """Add setup-specific arguments."""
+    
+    parser.add_argument("--check-deps", action="store_true", help="Check all dependencies")
+    parser.add_argument("--validate-env", action="store_true", help="Validate environment")
+    parser.add_argument("--setup-cuda", action="store_true", help="Setup CUDA environment")
+    parser.add_argument("--seed", type=int, default=42, help="Random seed for testing")
+
+
+def handle_train_command(args: argparse.Namespace) -> None:
+    """Handle training command."""
+    
+    logger = logging.getLogger("langvision.cli")
+    logger.info("Starting training...")
+    
+    # Initialize framework
+    quick_setup(seed=args.seed, log_level=args.log_level)
+    
+    # Create LoRA config if requested
+    lora_config = None
+    if args.use_lora:
+        lora_config = LoRAConfig(
+            r=args.lora_r,
+            alpha=args.lora_alpha,
+            dropout=args.lora_dropout
+        )
+        logger.info(f"Using LoRA with r={args.lora_r}, alpha={args.lora_alpha}")
+    
+    # Create model
+    if args.model == "vit":
+        model = VisionTransformer(
+            img_size=args.img_size,
+            num_classes=args.num_classes,
+            lora_config=lora_config
+        )
+    elif args.model.startswith("resnet"):
+        model_fn = {
+            "resnet18": "resnet18", "resnet34": "resnet34", "resnet50": "resnet50",
+            "resnet101": "resnet101", "resnet152": "resnet152"
+        }[args.model]
+        from ..models import resnet
+        model = getattr(resnet, model_fn)(num_classes=args.num_classes, lora_config=lora_config)
+    else:
+        raise ValueError(f"Unsupported model: {args.model}")
+    
+    logger.info(f"Created {args.model} model with {args.num_classes} classes")
+    
+    # Create dataset config
+    dataset_config = DatasetConfig(
+        root_dir=args.data_dir,
+        image_size=(args.img_size, args.img_size),
+        batch_size=args.batch_size,
+        num_workers=args.num_workers,
+        use_augmentation=args.augmentation,
+        augmentation_strength=args.augmentation_strength
+    )
+    
+    # Create dataloaders
+    dataloaders = create_enhanced_dataloaders(dataset_config)
+    logger.info(f"Created dataloaders for {args.dataset}")
+    
+    # Create training config
+    training_config = TrainingConfig(
+        epochs=args.epochs,
+        batch_size=args.batch_size,
+        learning_rate=args.lr,
+        weight_decay=args.weight_decay,
+        optimizer=args.optimizer,
+        scheduler=args.scheduler,
+        warmup_epochs=args.warmup_epochs,
+        use_amp=args.use_amp,
+        gradient_clip_norm=args.gradient_clip,
+        output_dir=args.output_dir,
+        experiment_name=args.experiment_name,
+        save_interval=args.save_interval,
+        log_interval=args.log_interval,
+        lora_config=lora_config,
+        freeze_backbone=args.freeze_backbone
+    )
+    
+    # Create trainer
+    trainer = AdvancedTrainer(
+        model=model,
+        train_loader=dataloaders['train'],
+        val_loader=dataloaders.get('val'),
+        config=training_config
+    )
+    
+    # Start training
+    trainer.train()
+    logger.info("Training completed!")
+
+
+def handle_setup_command(args: argparse.Namespace) -> None:
+    """Handle setup command."""
+    
+    config = {
+        "log_level": "INFO",
+        "seed": args.seed
+    }
+    
+    if args.setup_cuda:
+        config["setup_cuda"] = True
+    
+    results = initialize_langvision(config)
+    
+    print("\n=== Langvision Environment Setup ===")
+    print(f"Python: {results['validation_results']['python_version']}")
+    print(f"PyTorch: {results['validation_results']['pytorch_version']}")
+    
+    if results['validation_results']['cuda_available']:
+        print(f"CUDA: {results['validation_results']['cuda_version']}")
+        print(f"GPUs: {results['validation_results']['gpu_count']}")
+    else:
+        print("CUDA: Not available")
+    
+    if args.check_deps:
+        print("\n=== Dependencies ===")
+        for dep, available in results['dependencies'].items():
+            status = "✓" if available else "✗"
+            print(f"{status} {dep}")
+    
+    print("\nSetup completed successfully!")
+
+
+def main() -> None:
+    """Main CLI entry point."""
+    
+    parser = create_parser()
+    args = parser.parse_args()
+    
+    if not args.command:
+        parser.print_help()
+        sys.exit(1)
+    
+    try:
+        if args.command == "train":
+            handle_train_command(args)
+        elif args.command == "train-multimodal":
+            # TODO: Implement multimodal training handler
+            print("Multimodal training not yet implemented in CLI")
+            sys.exit(1)
+        elif args.command == "evaluate":
+            # TODO: Implement evaluation handler
+            print("Evaluation not yet implemented in CLI")
+            sys.exit(1)
+        elif args.command == "setup":
+            handle_setup_command(args)
+        else:
+            parser.print_help()
+            sys.exit(1)
+    
+    except Exception as e:
+        logger = logging.getLogger("langvision.cli")
+        logger.error(f"Command failed: {str(e)}")
+        sys.exit(1)
+
+
+if __name__ == "__main__":
+    main()