Upload .\scripts\train.py with huggingface_hub

.//scripts//train.py

@@ -0,0 +1,120 @@
+"""Main training script for BwengeAi."""
+
+import logging
+import sys
+from pathlib import Path
+
+import yaml
+
+PROJECT_ROOT = Path(__file__).resolve().parent.parent
+
+sys.path.insert(0, str(PROJECT_ROOT / "src"))
+
+from models.bwenge_model import BwengeModel
+from training.trainer import BwengeTrainer
+from evaluation.metrics import BwengeEvaluator
+
+logging.basicConfig(
+    level=logging.INFO,
+    format="%(asctime)s - %(name)s - %(levelname)s - %(message)s",
+)
+logger = logging.getLogger(__name__)
+
+
+def load_config(config_path: str = None) -> dict:
+    """Load configuration."""
+    if config_path is None:
+        config_path = str(PROJECT_ROOT / "configs/default.yaml")
+    with open(config_path, "r", encoding="utf-8") as f:
+        return yaml.safe_load(f)
+
+
+def main():
+    """Run training pipeline."""
+    logger.info("=" * 60)
+    logger.info("BwengeAi Training Pipeline")
+    logger.info("=" * 60)
+    
+    config = load_config()
+    
+    logger.info("\n" + "=" * 40)
+    logger.info("Step 1: Loading model")
+    logger.info("=" * 40)
+    
+    bwenge = BwengeModel(config)
+    
+    training_type = config.get("training", {}).get("training_type", "full")
+    
+    use_lora = training_type in ("lora", "qlora", "full")
+    use_quantization = training_type in ("qlora",)
+    
+    logger.info(f"Training type: {training_type} (LoRA={use_lora}, quantization={use_quantization})")
+    
+    model_name = config.get("model", {}).get("finetune", {}).get("base_models", [None])[0]
+    if not model_name:
+        model_name = config.get("model", {}).get("base_model", "meta-llama/Llama-3.2-1B")
+    
+    model, tokenizer = bwenge.load_base_model(
+        model_name=model_name,
+        use_quantization=use_quantization,
+    )
+    
+    if use_lora:
+        logger.info("Setting up LoRA for efficient fine-tuning...")
+        model = bwenge.setup_lora(model)
+    
+    logger.info("\n" + "=" * 40)
+    logger.info("Step 2: Preparing training data")
+    logger.info("=" * 40)
+    
+    trainer = BwengeTrainer(config)
+    
+    processed_dir = str(PROJECT_ROOT / config.get("data", {}).get("processed_dir", "data/processed"))
+    training_data_path = f"{processed_dir}/training_data.jsonl"
+    instruction_data_path = f"{processed_dir}/instruction_data.jsonl"
+    
+    if Path(instruction_data_path).exists():
+        data_path = instruction_data_path
+        logger.info(f"Using instruction dataset: {data_path}")
+    elif Path(training_data_path).exists():
+        data_path = training_data_path
+        logger.info(f"Using training dataset: {data_path}")
+    else:
+        logger.error("No training data found! Run collect_all.py first.")
+        return
+    
+    logger.info("\n" + "=" * 40)
+    logger.info("Step 3: Training")
+    logger.info("=" * 40)
+    
+    trainer.train_from_config(
+        model=model,
+        tokenizer=tokenizer,
+        data_path=data_path,
+        lora=use_lora,
+    )
+    
+    logger.info("\n" + "=" * 40)
+    logger.info("Step 4: Evaluation")
+    logger.info("=" * 40)
+    
+    evaluator = BwengeEvaluator(config)
+    
+    if Path(instruction_data_path).exists():
+        eval_results = evaluator.evaluate_model(
+            model=model,
+            tokenizer=tokenizer,
+            eval_data_path=instruction_data_path,
+            output_dir=str(trainer.output_dir),
+        )
+        logger.info(f"Evaluation results: {eval_results}")
+    
+    logger.info("\n" + "=" * 60)
+    logger.info("Training Pipeline Complete!")
+    logger.info("=" * 60)
+    
+    logger.info(f"Model saved to: {trainer.output_dir / 'final'}")
+
+
+if __name__ == "__main__":
+    main()