train.py

# train.py - Training script for Hugging Face

import os
import torch
import pandas as pd
import numpy as np
from datasets import load_dataset, DatasetDict
from transformers import (
    AutoTokenizer,
    AutoModelForSequenceClassification,
    TrainingArguments,
    Trainer,
    DataCollatorWithPadding
)
import evaluate
from sklearn.model_selection import train_test_split
import json

# Configuration
CONFIG = {
    "model_name": "distilbert-base-uncased",
    "dataset_name": "6StringNinja/synthetic-servicenow-incidents",
    "output_dir": "./incident-classifier",
    "test_size": 0.2,
    "random_state": 42,
    "max_length": 128,
    "batch_size": 16,
    "learning_rate": 2e-5,
    "num_epochs": 3,
    "push_to_hub": True,
    "hub_model_id": "Rajeshwartiwari/incident-classification-model"
}

def load_and_prepare_data():
    """Load and split dataset"""
    print("📂 Loading dataset...")
    dataset = load_dataset(CONFIG["dataset_name"])
    
    # Convert to pandas for splitting
    df = pd.DataFrame(dataset["train"])
    
    # Train/test split
    train_df, test_df = train_test_split(
        df,
        test_size=CONFIG["test_size"],
        random_state=CONFIG["random_state"],
        stratify=df["category"]
    )
    
    # Convert back to Hugging Face datasets
    train_dataset = DatasetDict({"train": dataset["train"].new_from_pandas(train_df)})
    test_dataset = DatasetDict({"test": dataset["train"].new_from_pandas(test_df)})
    
    # Get labels
    labels = sorted(list(set(df["category"])))
    label2id = {label: i for i, label in enumerate(labels)}
    id2label = {i: label for i, label in enumerate(labels)}
    
    print(f"✅ Dataset loaded. Categories: {labels}")
    print(f"   Train samples: {len(train_df)}, Test samples: {len(test_df)}")
    
    return train_dataset["train"], test_dataset["test"], label2id, id2label

def tokenize_function(examples, tokenizer):
    """Tokenize the examples"""
    # Combine short description and description
    texts = [
        f"{sd} {d}" if d else sd 
        for sd, d in zip(examples['short_description'], examples['description'])
    ]
    
    # Tokenize
    tokenized = tokenizer(
        texts,
        truncation=True,
        padding=True,
        max_length=CONFIG["max_length"]
    )
    
    # Add labels
    tokenized["labels"] = [label2id[l] for l in examples["category"]]
    return tokenized

def compute_metrics(eval_pred):
    """Compute evaluation metrics"""
    metric = evaluate.load("accuracy")
    logits, labels = eval_pred
    predictions = np.argmax(logits, axis=-1)
    
    # Calculate accuracy
    accuracy = metric.compute(predictions=predictions, references=labels)
    
    # You can add more metrics here
    return accuracy

def main():
    """Main training function"""
    print("🚀 Starting Incident Classification Model Training")
    
    # Load data
    train_dataset, test_dataset, label2id, id2label = load_and_prepare_data()
    
    # Initialize tokenizer and model
    print("🔧 Initializing tokenizer and model...")
    tokenizer = AutoTokenizer.from_pretrained(CONFIG["model_name"])
    model = AutoModelForSequenceClassification.from_pretrained(
        CONFIG["model_name"],
        num_labels=len(label2id),
        id2label=id2label,
        label2id=label2id
    )
    
    # Tokenize datasets
    print("🔠 Tokenizing datasets...")
    tokenized_train = train_dataset.map(
        lambda x: tokenize_function(x, tokenizer),
        batched=True,
        remove_columns=train_dataset.column_names
    )
    
    tokenized_test = test_dataset.map(
        lambda x: tokenize_function(x, tokenizer),
        batched=True,
        remove_columns=test_dataset.column_names
    )
    
    # Data collator
    data_collator = DataCollatorWithPadding(tokenizer=tokenizer)
    
    # Training arguments
    training_args = TrainingArguments(
        output_dir=CONFIG["output_dir"],
        learning_rate=CONFIG["learning_rate"],
        per_device_train_batch_size=CONFIG["batch_size"],
        per_device_eval_batch_size=CONFIG["batch_size"],
        num_train_epochs=CONFIG["num_epochs"],
        weight_decay=0.01,
        evaluation_strategy="epoch",
        save_strategy="epoch",
        load_best_model_at_end=True,
        metric_for_best_model="accuracy",
        report_to="none",  # Disable WandB by default
        push_to_hub=CONFIG["push_to_hub"],
        hub_model_id=CONFIG["hub_model_id"],
        hub_strategy="end",
        save_total_limit=2,
    )
    
    # Initialize trainer
    trainer = Trainer(
        model=model,
        args=training_args,
        train_dataset=tokenized_train,
        eval_dataset=tokenized_test,
        tokenizer=tokenizer,
        data_collator=data_collator,
        compute_metrics=compute_metrics,
    )
    
    # Train
    print("🎯 Starting training...")
    trainer.train()
    
    # Evaluate
    print("📊 Evaluating model...")
    eval_results = trainer.evaluate()
    print(f"✅ Evaluation results: {eval_results}")
    
    # Save everything locally
    print("💾 Saving model locally...")
    trainer.save_model(CONFIG["output_dir"])
    tokenizer.save_pretrained(CONFIG["output_dir"])
    
    # Save label mappings
    with open(os.path.join(CONFIG["output_dir"], "label_mappings.json"), "w") as f:
        json.dump({"label2id": label2id, "id2label": id2label}, f, indent=2)
    
    # Save configuration
    with open(os.path.join(CONFIG["output_dir"], "config.json"), "w") as f:
        json.dump(CONFIG, f, indent=2)
    
    print(f"🎉 Training complete! Model saved to {CONFIG['output_dir']}")
    
    if CONFIG["push_to_hub"]:
        print("☁️ Pushing to Hugging Face Hub...")
        trainer.push_to_hub()
        tokenizer.push_to_hub(CONFIG["hub_model_id"])
        print(f"✅ Model pushed to: https://huggingface.co/{CONFIG['hub_model_id']}")

if __name__ == "__main__":
    main()