#!/usr/bin/env python3
"""
Free Fine-Tuning Script for DualBlind Arena Reasoning Dataset
Runs on free Google Colab (T4 / L4 GPU) or local Linux/Mac with CUDA/MPS.
"""

import os
from datasets import load_dataset
from unsloth import FastLanguageModel
from trl import SFTTrainer
from transformers import TrainingArguments

DATASET_NAME = "GlimmaryKarl/DualBlind"
BASE_MODEL = "unsloth/Qwen2.5-7B-Instruct"  # Or "unsloth/Llama-3.1-8B-Instruct"

print(f"Loading base model: {BASE_MODEL}")
model, tokenizer = FastLanguageModel.from_pretrained(
    model_name=BASE_MODEL,
    max_seq_length=4096,
    load_in_4bit=True,
)

model = FastLanguageModel.get_peft_model(
    model,
    r=16,
    target_modules=["q_proj", "k_proj", "v_proj", "o_proj", "gate_proj", "up_proj", "down_proj"],
    lora_alpha=16,
    lora_dropout=0,
    bias="none",
    use_gradient_checkpointing="unsloth",
)

print(f"Loading reasoning SFT dataset from {DATASET_NAME}...")
try:
    dataset = load_dataset(DATASET_NAME, "sft_reasoning", split="train")
except Exception:
    print("Loading from local data/sft_reasoning_train.jsonl...")
    dataset = load_dataset("json", data_files="data/sft_reasoning_train.jsonl", split="train")

def format_prompts(batch):
    texts = []
    for msgs in batch["messages"]:
        texts.append(tokenizer.apply_chat_template(msgs, tokenize=False, add_generation_prompt=False))
    return {"text": texts}

dataset = dataset.map(format_prompts, batched=True)

trainer = SFTTrainer(
    model=model,
    tokenizer=tokenizer,
    train_dataset=dataset,
    dataset_text_field="text",
    max_seq_length=4096,
    dataset_num_proc=2,
    packing=False,
    args=TrainingArguments(
        per_device_train_batch_size=2,
        gradient_accumulation_steps=4,
        warmup_ratio=0.05,
        num_train_epochs=3,
        learning_rate=2e-4,
        fp16=True,
        logging_steps=5,
        optim="adamw_8bit",
        output_dir="reasoning_model_lora",
        save_strategy="epoch",
    ),
)

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

print("Saving fine-tuned LoRA weights...")
model.save_pretrained_merged("dualblind_reasoning_model", tokenizer, save_method="merged_16bit")
print("Complete! Model saved to ./dualblind_reasoning_model")