# Copyright 2020-2025 The HuggingFace Team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

# /// script
# dependencies = [
#     "trl @ git+https://github.com/huggingface/trl.git",
#     "trackio",
#     "kernels",
# ]
# ///

"""
Full training:
python examples/scripts/reward_modeling.py \
    --model_name_or_path Qwen/Qwen2-0.5B-Instruct \
    --dataset_name trl-lib/ultrafeedback_binarized \
    --output_dir Qwen2-0.5B-Reward \
    --per_device_train_batch_size 8 \
    --num_train_epochs 1 \
    --gradient_checkpointing True \
    --learning_rate 1.0e-5 \
    --eval_strategy steps \
    --eval_steps 50 \
    --max_length 2048

LoRA:
python examples/scripts/reward_modeling.py \
    --model_name_or_path Qwen/Qwen2-0.5B-Instruct \
    --dataset_name trl-lib/ultrafeedback_binarized \
    --output_dir Qwen2-0.5B-Reward-LoRA \
    --per_device_train_batch_size 8 \
    --num_train_epochs 1 \
    --gradient_checkpointing True \
    --learning_rate 1.0e-4 \
    --eval_strategy steps \
    --eval_steps 50 \
    --max_length 2048 \
    --use_peft \
    --lora_task_type SEQ_CLS \
    --lora_r 32 \
    --lora_alpha 16
"""

import os

import torch
from accelerate import logging
from datasets import load_dataset
from transformers import AutoModelForSequenceClassification, AutoTokenizer, HfArgumentParser

from trl import (
    ModelConfig,
    RewardConfig,
    RewardTrainer,
    ScriptArguments,
    get_kbit_device_map,
    get_peft_config,
    get_quantization_config,
    setup_chat_format,
)


logger = logging.get_logger(__name__)

# Enable logging in a Hugging Face Space
os.environ.setdefault("TRACKIO_SPACE_ID", "trl-trackio")


if __name__ == "__main__":
    parser = HfArgumentParser((ScriptArguments, RewardConfig, ModelConfig))
    script_args, training_args, model_args = parser.parse_args_into_dataclasses()
    training_args.gradient_checkpointing_kwargs = dict(use_reentrant=False)

    ################
    # Model & Tokenizer
    ################
    dtype = model_args.dtype if model_args.dtype in ["auto", None] else getattr(torch, model_args.dtype)
    model_kwargs = dict(
        revision=model_args.model_revision,
        use_cache=False if training_args.gradient_checkpointing else True,
        dtype=dtype,
    )
    quantization_config = get_quantization_config(model_args)
    if quantization_config is not None:
        # Passing None would not be treated the same as omitting the argument, so we include it only when valid.
        model_kwargs["device_map"] = get_kbit_device_map()
        model_kwargs["quantization_config"] = quantization_config

    tokenizer = AutoTokenizer.from_pretrained(
        model_args.model_name_or_path, trust_remote_code=model_args.trust_remote_code, use_fast=True
    )
    model = AutoModelForSequenceClassification.from_pretrained(
        model_args.model_name_or_path, num_labels=1, trust_remote_code=model_args.trust_remote_code, **model_kwargs
    )
    # Align padding tokens between tokenizer and model
    model.config.pad_token_id = tokenizer.pad_token_id

    # If post-training a base model, use ChatML as the default template
    if tokenizer.chat_template is None:
        model, tokenizer = setup_chat_format(model, tokenizer)

    if model_args.use_peft and model_args.lora_task_type != "SEQ_CLS":
        logger.warning(
            "You are using a `task_type` that is different than `SEQ_CLS` for PEFT. This will lead to silent bugs"
            " Make sure to pass --lora_task_type SEQ_CLS when using this script with PEFT.",
        )

    ##############
    # Load dataset
    ##############
    dataset = load_dataset(script_args.dataset_name, name=script_args.dataset_config)

    ##########
    # Training
    ##########
    trainer = RewardTrainer(
        model=model,
        processing_class=tokenizer,
        args=training_args,
        train_dataset=dataset[script_args.dataset_train_split],
        eval_dataset=dataset[script_args.dataset_test_split] if training_args.eval_strategy != "no" else None,
        peft_config=get_peft_config(model_args),
    )
    trainer.train()

    ############################
    # Save model and push to Hub
    ############################
    trainer.save_model(training_args.output_dir)

    if training_args.eval_strategy != "no":
        metrics = trainer.evaluate()
        trainer.log_metrics("eval", metrics)
        trainer.save_metrics("eval", metrics)

    # Save and push to hub
    trainer.save_model(training_args.output_dir)
    if training_args.push_to_hub:
        trainer.push_to_hub(dataset_name=script_args.dataset_name)