gpt-neox-20b.py

#!/usr/bin/env python
# coding=utf-8
# Copyright 2022 The HuggingFace Inc. 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.
"""
Fine-tuning the library models for causal language modeling (GPT, GPT-2, CTRL, ...)
on a text file or a dataset without using HuggingFace Trainer.

Here is the full list of checkpoints on the hub that can be fine-tuned by this script:
https://huggingface.co/models?filter=text-generation
"""
# You can also adapt this script on your own causal language modeling task. Pointers for this are left as comments.

# Adapted to work with Ray
# original script: https://github.com/huggingface/accelerate/blob/main/examples/by_feature/deepspeed_with_config_support.py

import ray

from ray.train.torch import TorchTrainer
from ray.air import ScalingConfig, RunConfig, session, Checkpoint
from ray.tune import SyncConfig

import argparse
import json
import logging
import math
import os
import random
from itertools import chain
import copy
from pathlib import Path

import torch
from torch.utils.data import DataLoader


def parse_args():
    parser = argparse.ArgumentParser(description="Finetune a transformers model on a causal language modeling task")
    parser.add_argument(
        "--dataset_name",
        type=str,
        default=None,
        help="The name of the dataset to use (via the datasets library).",
    )
    parser.add_argument(
        "--dataset_config_name",
        type=str,
        default=None,
        help="The configuration name of the dataset to use (via the datasets library).",
    )
    parser.add_argument(
        "--train_file", type=str, default=None, help="A csv or a json file containing the training data."
    )
    parser.add_argument(
        "--validation_file", type=str, default=None, help="A csv or a json file containing the validation data."
    )
    parser.add_argument(
        "--validation_split_percentage",
        default=5,
        help="The percentage of the train set used as validation set in case there's no validation split",
    )
    parser.add_argument(
        "--model_name_or_path",
        type=str,
        help="Path to pretrained model or model identifier from huggingface.co/models.",
        required=False,
    )
    parser.add_argument(
        "--config_name",
        type=str,
        default=None,
        help="Pretrained config name or path if not the same as model_name",
    )
    parser.add_argument(
        "--max_length",
        type=int,
        default=512,
        help="Maximum sequence length. Sequences will be right padded (and possibly truncated).",
    )
    parser.add_argument(
        "--tokenizer_name",
        type=str,
        default=None,
        help="Pretrained tokenizer name or path if not the same as model_name",
    )
    parser.add_argument(
        "--use_slow_tokenizer",
        action="store_true",
        help="If passed, will use a slow tokenizer (not backed by the 🤗 Tokenizers library).",
    )
    parser.add_argument(
        "--per_device_train_batch_size",
        type=int,
        default=8,
        help="Batch size (per device) for the training dataloader.",
    )
    parser.add_argument(
        "--per_device_eval_batch_size",
        type=int,
        default=8,
        help="Batch size (per device) for the evaluation dataloader.",
    )
    parser.add_argument(
        "--learning_rate",
        type=float,
        default=5e-5,
        help="Initial learning rate (after the potential warmup period) to use.",
    )
    parser.add_argument("--weight_decay", type=float, default=0.0, help="Weight decay to use.")
    parser.add_argument("--num_train_epochs", type=int, default=3, help="Total number of training epochs to perform.")
    parser.add_argument(
        "--max_train_steps",
        type=int,
        default=None,
        help="Total number of training steps to perform. If provided, overrides num_train_epochs.",
    )
    parser.add_argument(
        "--gradient_accumulation_steps",
        type=int,
        default=1,
        help="Number of updates steps to accumulate before performing a backward/update pass.",
    )
    parser.add_argument(
        "--lr_scheduler_type",
        default="linear",
        help="The scheduler type to use.",
        choices=["linear", "cosine", "cosine_with_restarts", "polynomial", "constant", "constant_with_warmup"],
    )
    parser.add_argument(
        "--num_warmup_steps", type=int, default=0, help="Number of steps for the warmup in the lr scheduler."
    )
    parser.add_argument("--output_dir", type=str, default=None, help="Where to store the final model.")
    parser.add_argument("--seed", type=int, default=None, help="A seed for reproducible training.")
    parser.add_argument(
        "--model_type",
        type=str,
        default=None,
        help="Model type to use if training from scratch.",
    )
    parser.add_argument(
        "--block_size",
        type=int,
        default=None,
        help=(
            "Optional input sequence length after tokenization. The training dataset will be truncated in block of"
            " this size for training. Default to the model max input length for single sentence inputs (take into"
            " account special tokens)."
        ),
    )
    parser.add_argument(
        "--preprocessing_num_workers",
        type=int,
        default=None,
        help="The number of processes to use for the preprocessing.",
    )
    parser.add_argument(
        "--overwrite_cache", type=bool, default=False, help="Overwrite the cached training and evaluation sets"
    )
    parser.add_argument(
        "--no_keep_linebreaks", action="store_true", help="Do not keep line breaks when using TXT files."
    )
    parser.add_argument("--push_to_hub", action="store_true", help="Whether or not to push the model to the Hub.")
    parser.add_argument(
        "--hub_model_id", type=str, help="The name of the repository to keep in sync with the local `output_dir`."
    )
    parser.add_argument("--hub_token", type=str, help="The token to use to push to the Model Hub.")
    parser.add_argument(
        "--checkpointing_steps",
        type=str,
        default=None,
        help="Whether the various states should be saved at the end of every n steps, or 'epoch' for each epoch.",
    )
    parser.add_argument(
        "--resume_from_checkpoint",
        type=str,
        default=None,
        help="If the training should continue from a checkpoint folder.",
    )
    # New Code #
    # Whether to load the best model at the end of training
    parser.add_argument(
        "--load_best_model",
        action="store_true",
        help="Whether to load the best model at the end of training",
    )
    parser.add_argument(
        "--with_tracking",
        action="store_true",
        help="Whether to enable experiment trackers for logging.",
    )
    parser.add_argument(
        "--report_to",
        type=str,
        default="all",
        help=(
            'The integration to report the results and logs to. Supported platforms are `"tensorboard"`,'
            ' `"wandb"` and `"comet_ml"`. Use `"all"` (default) to report to all integrations.'
            "Only applicable when `--with_tracking` is passed."
        ),
    )
    parser.add_argument(
        "--num_workers",
        type=int,
        default=32,
        help=(
            "num workers to use"
        ),
    )
    parser.add_argument(
        "--upload_dir",
        type=str,
        default=None,
        help=(
            "tune upload dir"
        ),
    )
    parser.add_argument(
        "--full_determinism",
        type=bool,
        default=False,
        help=(
            "disable torch determinism"
        ),
    )
    args = parser.parse_args()

    # Sanity checks
    if args.dataset_name is None and args.train_file is None and args.validation_file is None:
        raise ValueError("Need either a dataset name or a training/validation file.")
    else:
        if args.train_file is not None:
            extension = args.train_file.split(".")[-1]
            assert extension in ["csv", "json", "txt"], "`train_file` should be a csv, json or txt file."
        if args.validation_file is not None:
            extension = args.validation_file.split(".")[-1]
            assert extension in ["csv", "json", "txt"], "`validation_file` should be a csv, json or txt file."

    if args.push_to_hub:
        assert args.output_dir is not None, "Need an `output_dir` to create a repo when `--push_to_hub` is passed."

    return args


# New Code #
def checkpoint_model(checkpoint_folder, ckpt_id, model, epoch, last_global_step, **kwargs):
    """Utility function for checkpointing model + optimizer dictionaries
    The main purpose for this is to be able to resume training from that instant again
    """
    checkpoint_state_dict = {
        "epoch": epoch,
        "last_global_step": last_global_step,
    }
    # Add extra kwargs too
    checkpoint_state_dict.update(kwargs)

    success = model.save_checkpoint(checkpoint_folder, ckpt_id, checkpoint_state_dict)
    status_msg = f"checkpointing: checkpoint_folder={checkpoint_folder}, ckpt_id={ckpt_id}"
    if success:
        logging.info(f"Success {status_msg}")
    else:
        logging.warning(f"Failure {status_msg}")
    return


# New Code #
def load_training_checkpoint(model, load_dir, tag=None, **kwargs):
    """Utility function for checkpointing model + optimizer dictionaries
    The main purpose for this is to be able to resume training from that instant again
    """
    _, checkpoint_state_dict = model.load_checkpoint(load_dir, tag=tag, **kwargs)
    epoch = checkpoint_state_dict["epoch"]
    last_global_step = checkpoint_state_dict["last_global_step"]
    del checkpoint_state_dict
    return (epoch, last_global_step)


# New Code #
def evaluate(args, model, eval_dataloader, accelerator, eval_dataset):
    model.eval()
    losses = []
    for step, batch in enumerate(eval_dataloader):
        with torch.no_grad():
            outputs = model(**batch)

        loss = outputs.loss
        losses.append(accelerator.gather(loss.repeat(args.per_device_eval_batch_size)))

    losses = torch.cat(losses)
    losses = losses[: len(eval_dataset)]
    try:
        eval_loss = torch.mean(losses)
        perplexity = math.exp(eval_loss)
    except OverflowError:
        perplexity = float("inf")
    return perplexity, eval_loss


def training_loop(config):
    os.environ["HF_HOME"] = "/nvme/huggingface"

    import datasets
    import transformers
    from accelerate import Accelerator, DistributedType, DeepSpeedPlugin
    from accelerate.logging import get_logger
    from accelerate.utils import DummyOptim, DummyScheduler, set_seed
    from datasets import load_dataset
    from huggingface_hub import Repository
    from tqdm.auto import tqdm
    from transformers import (
        CONFIG_MAPPING,
        MODEL_MAPPING,
        AutoConfig,
        AutoModelForCausalLM,
        AutoTokenizer,
        SchedulerType,
        default_data_collator,
        get_scheduler,
    )
    from transformers.utils import get_full_repo_name
    from transformers.utils.versions import require_version

    # Env vars necessary for HF to setup DDP
    os.environ["RANK"] = str(session.get_world_rank())
    os.environ["WORLD_SIZE"] = str(session.get_world_size())
    os.environ["LOCAL_RANK"] = str(session.get_local_rank())
    os.environ["OMP_NUM_THREADS"] = str(
        session.get_trial_resources().bundles[-1].get("CPU", 1)
    )

    logger = get_logger(__name__)

    args = config["args"]
    args.lr_scheduler_type = SchedulerType(args.lr_scheduler_type)

    # Initialize the accelerator. We will let the accelerator handle device placement for us in this example.
    # If we're using tracking, we also need to initialize it here and it will by default pick up all supported trackers
    # in the environment
    deepspeed_plugin = DeepSpeedPlugin(
        hf_ds_config = {
            "fp16": {
                "enabled": False,
            },
            "bf16": {"enabled": True},
            "optimizer": {
                "type": "AdamW",
            },
            "scheduler": {
                "type": "WarmupLR",
                "params": {
                    "warmup_min_lr": "auto",
                    "warmup_max_lr": "auto",
                    "warmup_num_steps": "auto"
                }
            },
            "zero_optimization": {
                "stage": 3,
                # "offload_optimizer": {
                #     "device": "cpu",
                #     #"nvme_path": "/nvme",
                #  "pin_memory": False,
                # },
                # "offload_param": {
                #     "device": "cpu",
                #     #"nvme_path": "/nvme",
                #     "pin_memory": False,
                # },
                "reduce_bucket_size": "auto",
                "stage3_prefetch_bucket_size": "auto",
                "stage3_param_persistence_threshold": "auto",
                "gather_16bit_weights_on_model_save": True,
                "round_robin_gradients": True,
            },
            "gradient_accumulation_steps": "auto",
            "gradient_clipping": "auto",
            "steps_per_print": 1,
            "train_batch_size": "auto",
            "train_micro_batch_size_per_gpu": "auto",
            "wall_clock_breakdown": False,
        },
        zero3_init_flag=True,
    )
    accelerator = Accelerator(
        gradient_accumulation_steps=args.gradient_accumulation_steps,
        mixed_precision="bf16",
        deepspeed_plugin=deepspeed_plugin,
        #rng_types=["torch"],
    )
    # Make one log on every process with the configuration for debugging.
    logging.basicConfig(
        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
        datefmt="%m/%d/%Y %H:%M:%S",
        level=logging.INFO,
    )
    logger.info(accelerator.state, main_process_only=False)
    if accelerator.is_local_main_process:
        datasets.utils.logging.set_verbosity_warning()
        transformers.utils.logging.set_verbosity_info()
    else:
        datasets.utils.logging.set_verbosity_error()
        transformers.utils.logging.set_verbosity_error()

    # If passed along, set the training seed now.
    if args.seed is not None:
        set_seed(args.seed)

    # Handle the repository creation
    if accelerator.is_main_process:
        if args.push_to_hub:
            if args.hub_model_id is None:
                repo_name = get_full_repo_name(Path(args.output_dir).name, token=args.hub_token)
            else:
                repo_name = args.hub_model_id
            repo = Repository(args.output_dir, clone_from=repo_name)

            with open(os.path.join(args.output_dir, ".gitignore"), "w+") as gitignore:
                if "step_*" not in gitignore:
                    gitignore.write("step_*\n")
                if "epoch_*" not in gitignore:
                    gitignore.write("epoch_*\n")
        elif args.output_dir is not None:
            os.makedirs(args.output_dir, exist_ok=True)
    accelerator.wait_for_everyone()

    # Get the datasets: you can either provide your own CSV/JSON/TXT training and evaluation files (see below)
    # or just provide the name of one of the public datasets available on the hub at https://huggingface.co/datasets/
    # (the dataset will be downloaded automatically from the datasets Hub).
    #
    # For CSV/JSON files, this script will use the column called 'text' or the first column if no column called
    # 'text' is found. You can easily tweak this behavior (see below).
    #
    # In distributed training, the load_dataset function guarantee that only one local process can concurrently
    # download the dataset.
    if args.dataset_name is not None:
        # Downloading and loading a dataset from the hub.
        raw_datasets = load_dataset(args.dataset_name, args.dataset_config_name)
        if "validation" not in raw_datasets.keys():
            raw_datasets["validation"] = load_dataset(
                args.dataset_name,
                args.dataset_config_name,
                split=f"train[:{args.validation_split_percentage}%]",
            )
            raw_datasets["train"] = load_dataset(
                args.dataset_name,
                args.dataset_config_name,
                split=f"train[{args.validation_split_percentage}%:]",
            )
    else:
        raw_datasets = load_dataset(
            'csv', data_files={
                "train": args.train_file,
                "validation": args.validation_file,
            })

    # See more about loading any type of standard or custom dataset (from files, python dict, pandas DataFrame, etc) at
    # https://huggingface.co/docs/datasets/loading_datasets.html.

    # Load pretrained model and tokenizer
    #
    # In distributed training, the .from_pretrained methods guarantee that only one local process can concurrently
    # download model & vocab.
    if args.config_name:
        config = AutoConfig.from_pretrained(args.config_name, use_cache=False)
    elif args.model_name_or_path:
        config = AutoConfig.from_pretrained(args.model_name_or_path, use_cache=False)
    else:
        config = CONFIG_MAPPING[args.model_type]()
        logger.warning("You are instantiating a new config instance from scratch.")

    if args.tokenizer_name:
        tokenizer = AutoTokenizer.from_pretrained(args.tokenizer_name)
    elif args.model_name_or_path:
        tokenizer = AutoTokenizer.from_pretrained(args.model_name_or_path)
    else:
        raise ValueError(
            "You are instantiating a new tokenizer from scratch. This is not supported by this script."
            "You can do it from another script, save it, and load it from here, using --tokenizer_name."
        )

    if args.model_name_or_path:
        model = AutoModelForCausalLM.from_pretrained(
            args.model_name_or_path,
            from_tf=bool(".ckpt" in args.model_name_or_path),
            config=config,
            torch_dtype=torch.bfloat16,
        )
    else:
        logger.info("Training new model from scratch")
        model = AutoModelForCausalLM.from_config(config)

    model.resize_token_embeddings(len(tokenizer))

    # Preprocessing the datasets.

    def tokenize_function(examples):
        text_column = "question"
        label_column = "answer"
        instruction = "Below is an instruction that describes a task. Write a response that appropriately completes the request. ### Instruction: "
        response_prefix = "### Response: "
        inputs = [instruction + prompt + response_prefix + response+tokenizer.eos_token for prompt,response in zip(examples[text_column],examples[label_column])]

        model_inputs = tokenizer(inputs)
        model_inputs["labels"] = copy.deepcopy(model_inputs["input_ids"])
        return model_inputs

    with accelerator.main_process_first():
        tokenized_datasets = raw_datasets.map(
            tokenize_function,
            batched=True,
            num_proc=args.preprocessing_num_workers,
            remove_columns=raw_datasets["train"].column_names,
            load_from_cache_file=not args.overwrite_cache,
            desc="Running tokenizer on dataset",
        )

    if args.block_size is None:
        block_size = tokenizer.model_max_length
        if block_size > 1024:
            logger.warning(
                f"The tokenizer picked seems to have a very large `model_max_length` ({tokenizer.model_max_length}). "
                "Picking 1024 instead. You can change that default value by passing --block_size xxx."
            )
        block_size = 1024
    else:
        if args.block_size > tokenizer.model_max_length:
            logger.warning(
                f"The block_size passed ({args.block_size}) is larger than the maximum length for the model"
                f"({tokenizer.model_max_length}). Using block_size={tokenizer.model_max_length}."
            )
        block_size = min(args.block_size, tokenizer.model_max_length)

    # Main data processing function that will concatenate all texts from our dataset and generate chunks of block_size.
    def group_texts(examples):
        # Concatenate all texts.
        concatenated_examples = {k: list(chain(*examples[k])) for k in examples.keys()}
        total_length = len(concatenated_examples[list(examples.keys())[0]])
        # We drop the small remainder, we could add padding if the model supported it instead of this drop, you can
        # customize this part to your needs.
        if total_length >= block_size:
            total_length = (total_length // block_size) * block_size
        # Split by chunks of max_len.
        result = {
            k: [t[i : i + block_size] for i in range(0, total_length, block_size)]
            for k, t in concatenated_examples.items()
        }
        result["labels"] = result["input_ids"].copy()
        return result

    # Note that with `batched=True`, this map processes 1,000 texts together, so group_texts throws away a remainder
    # for each of those groups of 1,000 texts. You can adjust that batch_size here but a higher value might be slower
    # to preprocess.
    #
    # To speed up this part, we use multiprocessing. See the documentation of the map method for more information:
    # https://huggingface.co/docs/datasets/package_reference/main_classes.html#datasets.Dataset.map

    with accelerator.main_process_first():
        lm_datasets = tokenized_datasets.map(
            group_texts,
            batched=True,
            num_proc=args.preprocessing_num_workers,
            load_from_cache_file=not args.overwrite_cache,
            desc=f"Grouping texts in chunks of {block_size}",
        )

    accelerator.wait_for_everyone()
    train_dataset = lm_datasets["train"]
    eval_dataset = lm_datasets["validation"]

    # Log a few random samples from the training set:
    for index in random.sample(range(len(train_dataset)), 3):
        logger.info(f"Sample {index} of the training set: {train_dataset[index]}.")

    # DataLoaders creation:
    train_dataloader = DataLoader(
        train_dataset, shuffle=False, collate_fn=default_data_collator, batch_size=args.per_device_train_batch_size
    )
    eval_dataloader = DataLoader(
        eval_dataset, collate_fn=default_data_collator, batch_size=args.per_device_eval_batch_size
    )

    # Optimizer
    # Split weights in two groups, one with weight decay and the other not.
    no_decay = ["bias", "LayerNorm.weight"]
    optimizer_grouped_parameters = [
        {
            "params": [p for n, p in model.named_parameters() if not any(nd in n for nd in no_decay)],
            "weight_decay": args.weight_decay,
        },
        {
            "params": [p for n, p in model.named_parameters() if any(nd in n for nd in no_decay)],
            "weight_decay": 0.0,
        },
    ]
    # New Code #
    # Creates Dummy Optimizer if `optimizer` was specified in the config file else creates Adam Optimizer
    optimizer_cls = (
        torch.optim.AdamW
        if accelerator.state.deepspeed_plugin is None
        or "optimizer" not in accelerator.state.deepspeed_plugin.deepspeed_config
        else DummyOptim
    )
    optimizer = optimizer_cls(optimizer_grouped_parameters, lr=args.learning_rate)

    # On TPU, the tie weights in our model have been disconnected, so we need to restore the ties.
    if accelerator.distributed_type == DistributedType.TPU:
        model.tie_weights()

    # Scheduler and math around the number of training steps.

    # New Code
    # Get gradient accumulation steps from deepspeed config if available
    # if accelerator.state.deepspeed_plugin is not None:
    #     args.gradient_accumulation_steps = accelerator.state.deepspeed_plugin.deepspeed_config[
    #         "gradient_accumulation_steps"
    #     ]

    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
    if args.max_train_steps is None:
        args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
    else:
        args.num_train_epochs = math.ceil(args.max_train_steps / num_update_steps_per_epoch)

    # New Code #
    # Creates Dummy Scheduler if `scheduler` was specified in the config file else creates `args.lr_scheduler_type` Scheduler
    if (
        accelerator.state.deepspeed_plugin is None
        or "scheduler" not in accelerator.state.deepspeed_plugin.deepspeed_config
    ):
        lr_scheduler = get_scheduler(
            name=args.lr_scheduler_type,
            optimizer=optimizer,
            num_warmup_steps=args.num_warmup_steps,
            num_training_steps=args.max_train_steps,
        )
    else:
        lr_scheduler = DummyScheduler(
            optimizer, total_num_steps=args.max_train_steps, warmup_num_steps=args.num_warmup_steps
        )

    model.gradient_checkpointing_enable()

    # Prepare everything with our `accelerator`.
    model, optimizer, train_dataloader, eval_dataloader, lr_scheduler = accelerator.prepare(
        model, optimizer, train_dataloader, eval_dataloader, lr_scheduler
    )

    # We need to recalculate our total training steps as the size of the training dataloader may have changed.
    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
    args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch

    # Figure out how many steps we should save the Accelerator states
    if hasattr(args.checkpointing_steps, "isdigit"):
        checkpointing_steps = args.checkpointing_steps
        if args.checkpointing_steps.isdigit():
            checkpointing_steps = int(args.checkpointing_steps)
    else:
        checkpointing_steps = None

    # We need to initialize the trackers we use, and also store our configuration.
    # The trackers initializes automatically on the main process.
    if args.with_tracking:
        experiment_config = vars(args)
        # TensorBoard cannot log Enums, need the raw value
        experiment_config["lr_scheduler_type"] = experiment_config["lr_scheduler_type"].value
        accelerator.init_trackers("clm_no_trainer", experiment_config)

    # Train!
    total_batch_size = args.per_device_train_batch_size * accelerator.num_processes * args.gradient_accumulation_steps

    logger.info("***** Running training *****")
    logger.info(f"  Num examples = {len(train_dataset)}")
    logger.info(f"  Num Epochs = {args.num_train_epochs}")
    logger.info(f"  Instantaneous batch size per device = {args.per_device_train_batch_size}")
    logger.info(f"  Total train batch size (w. parallel, distributed & accumulation) = {total_batch_size}")
    logger.info(f"  Gradient Accumulation steps = {args.gradient_accumulation_steps}")
    logger.info(f"  Total optimization steps = {args.max_train_steps}")
    # Only show the progress bar once on each machine.
    progress_bar = tqdm(range(args.max_train_steps), disable=not accelerator.is_local_main_process)
    completed_steps = 0
    starting_epoch = 0
    best_metric = None
    best_metric_checkpoint = None

    accelerator.wait_for_everyone()
    # Potentially load in the weights and states from a previous save
    if session.get_checkpoint():
        with session.get_checkpoint().as_directory() as checkpoint_dir:
            # New Code #
            # Loads the DeepSpeed checkpoint from the specified path
            _, last_global_step = load_training_checkpoint(
                model,
                checkpoint_dir,
                **{"load_optimizer_states": True, "load_lr_scheduler_states": True},
            )
            accelerator.print(f"Resumed from checkpoint: {checkpoint_dir}")
            resume_step = last_global_step
            starting_epoch = resume_step // len(train_dataloader)
            resume_step -= starting_epoch * len(train_dataloader)

    checkpoint = None
    print("Starting training")
    for epoch in range(starting_epoch, args.num_train_epochs):
        model.train()
        total_loss = 0
        for step, batch in enumerate(train_dataloader):
            # We need to skip steps until we reach the resumed step
            if args.resume_from_checkpoint and epoch == starting_epoch:
                if resume_step is not None and step < resume_step:
                    completed_steps += 1
                    continue

            outputs = model(**batch)
            loss = outputs.loss
            # We keep track of the loss at each epoch
            total_loss += loss.detach().float()
            loss = loss / args.gradient_accumulation_steps
            accelerator.backward(loss)
            if step % args.gradient_accumulation_steps == 0 or step == len(train_dataloader) - 1:
                optimizer.step()
                lr_scheduler.step()
                optimizer.zero_grad()
                progress_bar.update(1)
                completed_steps += 1

            if isinstance(checkpointing_steps, int):
                if completed_steps % checkpointing_steps == 0:
                    output_dir = f"step_{completed_steps }"
                    if args.output_dir is not None:
                        output_dir = os.path.join(args.output_dir, output_dir)
                    accelerator.wait_for_everyone()
                    unwrapped_model = accelerator.unwrap_model(model)
                    unwrapped_model.save_pretrained(
                        output_dir,
                        is_main_process=accelerator.is_main_process,
                        save_function=accelerator.save,
                        state_dict=accelerator.get_state_dict(model),
                    )
                    if accelerator.is_main_process:
                        tokenizer.save_pretrained(output_dir)

            print(f"Completed steps: {completed_steps}/{args.max_train_steps}")
            if completed_steps >= args.max_train_steps:
                break

        perplexity, eval_loss = evaluate(args, model, eval_dataloader, accelerator, eval_dataset)
        logger.info(f"epoch {epoch}: perplexity: {perplexity} eval_loss: {eval_loss}")

        # New Code #
        # Save the DeepSpeed checkpoint to the specified path
        checkpoint_model(args.output_dir, epoch, model, epoch, completed_steps)

        checkpoint = Checkpoint.from_directory(args.output_dir)
        session.report(
            {
                "perplexity": perplexity,
                "eval_loss": eval_loss.float().cpu().numpy(),
                "train_loss": (total_loss.item() / len(train_dataloader)),
                "epoch": epoch,
                "step": completed_steps,
            },
            checkpoint=checkpoint
        )
        checkpoint = None

        # New Code #
        # Tracks the best checkpoint and best metric
        if best_metric is None or best_metric > perplexity:
            best_metric = perplexity
            best_metric_checkpoint = os.path.join(args.output_dir, str(epoch))
            accelerator.print(f"New best metric: {best_metric} at epoch {epoch}")
            accelerator.print(f"best_metric_checkpoint: {best_metric_checkpoint}")

        print(f"Completed epochs: {epoch+1}/{args.num_train_epochs}")

    # New Code #
    # Evaluates using the best checkpoint
    perplexity, eval_loss = evaluate(args, model, eval_dataloader, accelerator, eval_dataset)
    logger.info(f"Best model metrics: perplexity: {perplexity} eval_loss: {eval_loss}")
    if perplexity != best_metric:
        raise AssertionError(
            f"Best metric {best_metric} does not match the metric {perplexity} of the loaded best model."
        )

    if args.output_dir is not None:
        accelerator.wait_for_everyone()
        unwrapped_model = accelerator.unwrap_model(model)

        # New Code #
        # Saves the whole/unpartitioned fp16 model when in ZeRO Stage-3 to the output directory if
        # `stage3_gather_16bit_weights_on_model_save` is True in DeepSpeed Config file or
        # `zero3_save_16bit_model` is True in DeepSpeed Plugin.
        # For Zero Stages 1 and 2, models are saved as usual in the output directory.
        # The model name saved is `pytorch_model.bin`
        unwrapped_model.save_pretrained(
            args.output_dir,
            is_main_process=accelerator.is_main_process,
            save_function=accelerator.save,
            state_dict=accelerator.get_state_dict(model),
        )
        if accelerator.is_main_process:
            tokenizer.save_pretrained(args.output_dir)
            if args.push_to_hub:
                repo.push_to_hub(commit_message="End of training", auto_lfs_prune=True)

        checkpoint = Checkpoint.from_directory(args.output_dir)
        session.report(
            {
                "perplexity": perplexity,
                "eval_loss": eval_loss.float().cpu().numpy(),
                "train_loss": (total_loss.item() / len(train_dataloader)),
                "epoch": epoch,
                "step": completed_steps,
            },
            checkpoint=checkpoint
        )

        with open(os.path.join(args.output_dir, "all_results.json"), "w") as f:
            json.dump({"perplexity": perplexity, "eval_loss": eval_loss.item()}, f)


if __name__ == "__main__":
    #ray.init(runtime_env={"env_vars": {"NCCL_SOCKET_IFNAME": "ens5", "RAY_memory_monitor_refresh_ms": "0"}})
    ray.init()


    args = parse_args()
    if args.resume_from_checkpoint:
        checkpoint = Checkpoint.from_uri(args.resume_from_checkpoint)
    else:
        checkpoint = None

    trainer = TorchTrainer(
        training_loop,
        train_loop_config={"args": args},
        scaling_config=ScalingConfig(
            use_gpu=True, num_workers=args.num_workers, resources_per_worker={"GPU": 1, "CPU": 4}
        ),
        run_config=RunConfig(
            local_dir="/nvme/ray_results", sync_config=SyncConfig(upload_dir=args.upload_dir)
        ),
        resume_from_checkpoint=checkpoint,
    )
    result = trainer.fit()
    print(result)
    print(result.metrics_dataframe)
    print(result.checkpoint)