#!/usr/bin/env python3
# /// script
# dependencies = [
#     "trl>=0.12.0",
#     "peft>=0.7.0",
#     "transformers>=4.36.0",
#     "accelerate>=0.24.0",
#     "trackio",
# ]
# ///

"""
SFT Training Script for Qwen/Qwen2.5-0.5B

This script fine-tunes Qwen/Qwen2.5-0.5B using Supervised Fine-Tuning (SFT)
with LoRA for efficient training on the Capybara dataset.

Features:
- Trackio integration for real-time monitoring
- LoRA/PEFT for memory-efficient training
- Automatic Hub saving with checkpoints
- Train/eval split for progress monitoring
"""

import trackio
from datasets import load_dataset
from peft import LoraConfig
from trl import SFTTrainer, SFTConfig

# Load dataset
print("📦 Loading dataset...")
dataset = load_dataset("trl-lib/Capybara", split="train")
print(f"✅ Dataset loaded: {len(dataset)} examples")

# Create train/eval split
print("🔀 Creating train/eval split...")
dataset_split = dataset.train_test_split(test_size=0.1, seed=42)
train_dataset = dataset_split["train"]
eval_dataset = dataset_split["test"]
print(f"   Train: {len(train_dataset)} examples")
print(f"   Eval: {len(eval_dataset)} examples")

# Training configuration
config = SFTConfig(
    # CRITICAL: Hub settings
    output_dir="qwen-capybara-sft",
    push_to_hub=True,
    hub_model_id="vgtomahawk/qwen-capybara-sft",
    hub_strategy="every_save",  # Push checkpoints

    # Training parameters
    num_train_epochs=3,
    per_device_train_batch_size=4,
    gradient_accumulation_steps=4,
    learning_rate=2e-5,

    # Logging & checkpointing
    logging_steps=10,
    save_strategy="steps",
    save_steps=100,
    save_total_limit=2,

    # Evaluation
    eval_strategy="steps",
    eval_steps=100,

    # Optimization
    warmup_ratio=0.1,
    lr_scheduler_type="cosine",

    # Monitoring with Trackio
    report_to="trackio",
    project="qwen-sft-training",
    run_name="qwen-0.5b-capybara-baseline",
)

# LoRA configuration for efficient training
peft_config = LoraConfig(
    r=16,
    lora_alpha=32,
    lora_dropout=0.05,
    bias="none",
    task_type="CAUSAL_LM",
    target_modules=["q_proj", "v_proj"],
)

# Initialize trainer
print("🎯 Initializing trainer...")
trainer = SFTTrainer(
    model="Qwen/Qwen2.5-0.5B",
    train_dataset=train_dataset,
    eval_dataset=eval_dataset,
    args=config,
    peft_config=peft_config,
)

print("🚀 Starting training...")
trainer.train()

print("💾 Pushing final model to Hub...")
trainer.push_to_hub()

print("✅ Training complete!")
print(f"📦 Model: https://huggingface.co/vgtomahawk/qwen-capybara-sft")
print(f"📊 Metrics: https://huggingface.co/spaces/vgtomahawk/trackio")