tools/finetune_sts_v4_48_3.py

#!/usr/bin/env python
# coding=utf-8

import logging
import math
import os
import sys
from dataclasses import dataclass, field
from itertools import chain
from typing import Optional

import datasets
import evaluate
import torch
from datasets import load_dataset

from deepspeed import zero
from deepspeed.runtime.zero.partition_parameters import ZeroParamStatus
from accelerate import infer_auto_device_map, init_empty_weights
import transformers
from transformers import (
    CONFIG_MAPPING,
    MODEL_FOR_CAUSAL_LM_MAPPING,
    AutoConfig,
    AutoModelForCausalLM,
    AutoTokenizer,
    HfArgumentParser,
    # Trainer,
    # TrainingArguments,
    default_data_collator,
    is_torch_xla_available,
    set_seed,
)
from transformers import AutoModel, AutoConfig
from transformers.testing_utils import CaptureLogger
from transformers.trainer_utils import get_last_checkpoint
from transformers.utils import check_min_version, send_example_telemetry
from transformers.utils.versions import require_version
from trainer_v4_48_3 import Trainer


import vita_audio.models
from vita_audio import build_supervised_dataset_deepspeed
from vita_audio.tokenizer import update_tokenizer_for_s2s, get_audio_tokenizer


logger = logging.getLogger(__name__)


MODEL_CONFIG_CLASSES = list(MODEL_FOR_CAUSAL_LM_MAPPING.keys())
MODEL_TYPES = tuple(conf.model_type for conf in MODEL_CONFIG_CLASSES)


@dataclass
class ModelArguments:
    """
    Arguments pertaining to which model/config/tokenizer we are going to fine-tune, or train from scratch.
    """

    model_name_or_path: Optional[str] = field(
        default=None,
        metadata={
            "help": (
                "The model checkpoint for weights initialization. Don't set if you want to train a model from scratch."
            )
        },
    )
    model_type: Optional[str] = field(
        default=None,
        metadata={"help": "If training from scratch, pass a model type from the list: " + ", ".join(MODEL_TYPES)},
    )
    config_overrides: Optional[str] = field(
        default=None,
        metadata={
            "help": (
                "Override some existing default config settings when a model is trained from scratch. Example: "
                "n_embd=10,resid_pdrop=0.2,scale_attn_weights=false,summary_type=cls_index"
            )
        },
    )
    config_name: Optional[str] = field(
        default=None, metadata={"help": "Pretrained config name or path if not the same as model_name"}
    )
    tokenizer_name: Optional[str] = field(
        default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"}
    )
    cache_dir: Optional[str] = field(
        default=None,
        metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
    )
    use_fast_tokenizer: bool = field(
        default=True,
        metadata={"help": "Whether to use one of the fast tokenizer (backed by the tokenizers library) or not."},
    )
    model_revision: str = field(
        default="main",
        metadata={"help": "The specific model version to use (can be a branch name, tag name or commit id)."},
    )
    token: str = field(
        default=None,
        metadata={
            "help": (
                "The token to use as HTTP bearer authorization for remote files. If not specified, will use the token "
                "generated when running `huggingface-cli login` (stored in `~/.huggingface`)."
            )
        },
    )
    trust_remote_code: bool = field(
        default=False,
        metadata={
            "help": (
                "Whether to trust the execution of code from datasets/models defined on the Hub."
                " This option should only be set to `True` for repositories you trust and in which you have read the"
                " code, as it will execute code present on the Hub on your local machine."
            )
        },
    )
    torch_dtype: Optional[str] = field(
        default=None,
        metadata={
            "help": (
                "Override the default `torch.dtype` and load the model under this dtype. If `auto` is passed, the "
                "dtype will be automatically derived from the model's weights."
            ),
            "choices": ["auto", "bfloat16", "float16", "float32"],
        },
    )
    low_cpu_mem_usage: bool = field(
        default=False,
        metadata={
            "help": (
                "It is an option to create the model as an empty shell, then only materialize its parameters when the pretrained weights are loaded. "
                "set True will benefit LLM loading time and RAM consumption."
            )
        },
    )

    attn_implementation: Optional[str] = field(default=None, metadata={"help": ""})

    audio_tokenizer_path: str = field(default=None, metadata={"help": ""})
    audio_tokenizer_type: str = field(default=None, metadata={"help": ""})
    text_audio_interval_ratio: list[int] = field(default=None, metadata={"help": ""})
    audio_model_freeze: bool = field(default=False, metadata={"help": ""})

    vision_model_name_or_path: str = field(default=None, metadata={"help": ""})
    vision_model_type: Optional[str] = field(default=None, metadata={"help": ""})
    vision_model_freeze: bool = field(default=False, metadata={"help": ""})

    language_model_freeze: bool = field(default=False, metadata={"help": ""})

    vision_projector_type: str = field(default="mlp", metadata={"help": ""})
    vision_projector_pre_norm: bool = field(default=False, metadata={"help": ""})
    vision_downsample_ratio: float = field(default=0.5, metadata={"help": ""})

    image_size: int = field(default=448, metadata={"help": ""})
    image_token_length: int = field(default=1025, metadata={"help": ""})
    max_num_frame: int = field(default=16, metadata={"help": ""})
    max_fps: int = field(default=1, metadata={"help": ""})
    min_patch_grid: int = field(default=1, metadata={"help": ""})
    max_patch_grid: int = field(default=12, metadata={"help": ""})
    vision_process_type: str = field(default="dynamic", metadata={"help": ""})
    vision_normalize_type: str = field(default="imagenet", metadata={"help": ""})

    model_max_length: int = field(default=4096, metadata={"help": ""})

    A: int = field(default=0, metadata={"help": ""})
    B: str = field(default=None, metadata={"help": ""})
    C: bool = field(default=False, metadata={"help": ""})

    def __post_init__(self):
        if self.config_overrides is not None and (self.config_name is not None or self.model_name_or_path is not None):
            raise ValueError(
                "--config_overrides can't be used in combination with --config_name or --model_name_or_path"
            )


@dataclass
class DataTrainingArguments:
    """
    Arguments pertaining to what data we are going to input our model for training and eval.
    """

    dataset_name: Optional[str] = field(
        default=None, metadata={"help": "The name of the dataset to use (via the datasets library)."}
    )
    dataset_config_name: Optional[str] = field(
        default=None, metadata={"help": "The configuration name of the dataset to use (via the datasets library)."}
    )
    train_file: Optional[str] = field(default=None, metadata={"help": "The input training data file (a text file)."})
    validation_file: Optional[str] = field(
        default=None,
        metadata={"help": "An optional input evaluation data file to evaluate the perplexity on (a text file)."},
    )
    max_train_samples: Optional[int] = field(
        default=None,
        metadata={
            "help": (
                "For debugging purposes or quicker training, truncate the number of training examples to this "
                "value if set."
            )
        },
    )
    max_eval_samples: Optional[int] = field(
        default=None,
        metadata={
            "help": (
                "For debugging purposes or quicker training, truncate the number of evaluation examples to this "
                "value if set."
            )
        },
    )
    streaming: bool = field(default=False, metadata={"help": "Enable streaming mode"})
    block_size: Optional[int] = field(
        default=None,
        metadata={
            "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)."
            )
        },
    )
    overwrite_cache: bool = field(
        default=False, metadata={"help": "Overwrite the cached training and evaluation sets"}
    )
    validation_split_percentage: Optional[int] = field(
        default=5,
        metadata={
            "help": "The percentage of the train set used as validation set in case there's no validation split"
        },
    )
    preprocessing_num_workers: Optional[int] = field(
        default=None,
        metadata={"help": "The number of processes to use for the preprocessing."},
    )
    keep_linebreaks: bool = field(
        default=True, metadata={"help": "Whether to keep line breaks when using TXT files or not."}
    )

    create_attention_mask: bool = field(default=True, metadata={"help": "create_attention_mask"})
    create_attention_mask_2d: bool = field(default=False, metadata={"help": "create_attention_mask_2d"})
    reset_position_ids: bool = field(default=False, metadata={"help": ""})
    reset_attention_mask: bool = field(default=False, metadata={"help": ""})
    cross_dataset_joint: bool = field(default=False, metadata={"help": ""})

    # dataset_joint: bool = field(default=False, metadata={"help": ""})
    # variable_length: bool = field(default=True, metadata={"help": ""})
    dataset_joint: bool = field(default=True, metadata={"help": ""})
    variable_length: bool = field(default=False, metadata={"help": ""})

    D: int = field(default=0, metadata={"help": ""})
    E: str = field(default=None, metadata={"help": ""})
    F: bool = field(default=False, metadata={"help": ""})

    def __post_init__(self):
        if self.streaming:
            require_version("datasets>=2.0.0", "The streaming feature requires `datasets>=2.0.0`")

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


@dataclass
class TrainingArguments(transformers.TrainingArguments):
    vision_model_lr_mult: float = field(default=1.0, metadata={"help": ""})
    vision_model_lr_decay_rate: float = field(default=1.0, metadata={"help": ""})

    mtp_model_lr_mult: float = field(default=1.0, metadata={"help": ""})


def main():
    # See all possible arguments in src/transformers/training_args.py
    # or by passing the --help flag to this script.
    # We now keep distinct sets of args, for a cleaner separation of concerns.

    parser = HfArgumentParser((ModelArguments, DataTrainingArguments, TrainingArguments))
    if len(sys.argv) == 2 and sys.argv[1].endswith(".json"):
        # If we pass only one argument to the script and it's the path to a json file,
        # let's parse it to get our arguments.
        model_args, data_args, training_args = parser.parse_json_file(json_file=os.path.abspath(sys.argv[1]))
    else:
        model_args, data_args, training_args = parser.parse_args_into_dataclasses()

    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
    # information sent is the one passed as arguments along with your Python/PyTorch versions.
    send_example_telemetry("run_clm", model_args, data_args)

    # Setup logging
    logging.basicConfig(
        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
        datefmt="%m/%d/%Y %H:%M:%S",
        handlers=[logging.StreamHandler(sys.stdout)],
    )

    if training_args.should_log:
        # The default of training_args.log_level is passive, so we set log level at info here to have that default.
        transformers.utils.logging.set_verbosity_info()

    log_level = training_args.get_process_log_level()
    logger.setLevel(log_level)
    datasets.utils.logging.set_verbosity(log_level)
    transformers.utils.logging.set_verbosity(log_level)
    transformers.utils.logging.enable_default_handler()
    transformers.utils.logging.enable_explicit_format()

    # test_npu(training_args)
    # torch_npu.npu.set_device("npu")
    training_args.gradient_checkpointing_kwargs = {'use_reentrant': False}

    # Log on each process the small summary:
    logger.warning(
        f"Process rank: {training_args.local_rank}, device: {training_args.device}, n_gpu: {training_args.n_gpu}, "
        + f"distributed training: {training_args.parallel_mode.value == 'distributed'}, 16-bits training: {training_args.fp16}"
    )
    logger.info(f"Training/evaluation parameters {training_args}")

    # Detecting last checkpoint.
    last_checkpoint = None
    if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
        last_checkpoint = get_last_checkpoint(training_args.output_dir)
        if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
            raise ValueError(
                f"Output directory ({training_args.output_dir}) already exists and is not empty. "
                "Use --overwrite_output_dir to overcome."
            )
        elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
            logger.info(
                f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
                "the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
            )

    # Set seed before initializing model.
    set_seed(training_args.seed)

    # 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.

    # Load pretrained model and tokenizer
    #
    # Distributed training:
    # The .from_pretrained methods guarantee that only one local process can concurrently
    # download model & vocab.

    logger.info(f"{model_args=}")

    config_kwargs = {
        "cache_dir": model_args.cache_dir,
        "revision": model_args.model_revision,
        "token": model_args.token,
        "trust_remote_code": model_args.trust_remote_code,
        "model_max_length": model_args.model_max_length,
        "attn_implementation": model_args.attn_implementation,
    }
    if model_args.config_name:
        config = AutoConfig.from_pretrained(model_args.config_name, **config_kwargs)
    elif model_args.model_name_or_path:
        config = AutoConfig.from_pretrained(model_args.model_name_or_path, **config_kwargs)
    else:
        config = CONFIG_MAPPING[model_args.model_type]()
        logger.warning("You are instantiating a new config instance from scratch.")
        if model_args.config_overrides is not None:
            logger.info(f"Overriding config: {model_args.config_overrides}")
            config.update_from_string(model_args.config_overrides)
            logger.info(f"New config: {config}")
    config.use_cache = False
    logger.info(f"{config.__class__.__name__=} {config=}")

    tokenizer_kwargs = {
        "cache_dir": model_args.cache_dir,
        "use_fast": model_args.use_fast_tokenizer,
        "revision": model_args.model_revision,
        "token": model_args.token,
        "trust_remote_code": model_args.trust_remote_code,
    }
    if model_args.tokenizer_name:
        tokenizer = AutoTokenizer.from_pretrained(model_args.tokenizer_name, **tokenizer_kwargs)
    elif model_args.model_name_or_path:
        tokenizer = AutoTokenizer.from_pretrained(model_args.model_name_or_path, **tokenizer_kwargs)
    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 tokenizer.pad_token_id is None:
        if tokenizer.pad_id is not None:
            tokenizer.pad_token_id = tokenizer.pad_id
        else:
            tokenizer.pad_token_id = tokenizer.eod_id

    logger.info(f"{tokenizer.__class__.__name__=} {len(tokenizer)=}")
    if "HYTokenizer" in tokenizer.__class__.__name__:
        pass
    else:
        tokenizer = update_tokenizer_for_s2s(tokenizer, model_args.audio_tokenizer_type)
    logger.info(f"{tokenizer.__class__.__name__=} {len(tokenizer)=}")

    if model_args.model_name_or_path:
        torch_dtype = (
            model_args.torch_dtype
            if model_args.torch_dtype in ["auto", None]
            else getattr(torch, model_args.torch_dtype)
        )
        model, loading_info = AutoModelForCausalLM.from_pretrained(
            model_args.model_name_or_path,
            from_tf=bool(".ckpt" in model_args.model_name_or_path),
            config=config,
            cache_dir=model_args.cache_dir,
            revision=model_args.model_revision,
            token=model_args.token,
            trust_remote_code=model_args.trust_remote_code,
            torch_dtype=torch_dtype,
            low_cpu_mem_usage=model_args.low_cpu_mem_usage,
            output_loading_info=True
        )
    else:
        model = AutoModelForCausalLM.from_config(config, trust_remote_code=model_args.trust_remote_code)
        n_params = sum({p.data_ptr(): p.numel() for p in model.parameters()}.values())
        logger.info(f"Training new model from scratch - Total size={n_params/2**20:.2f}M params")

    if model_args.text_audio_interval_ratio is not None:
        model.generation_config.mtp_inference_mode = model_args.text_audio_interval_ratio

    if model_args.vision_model_name_or_path:
        logger.info(f"Loading {model_args.vision_model_name_or_path}")
        torch_dtype = (
            model_args.torch_dtype
            if model_args.torch_dtype in ["auto", None]
            else getattr(torch, model_args.torch_dtype)
        )
        vision_config = AutoConfig.from_pretrained(
            model_args.vision_model_name_or_path,
            trust_remote_code=True,
        )
        vision_config.attention_dropout = 0.0
        vision_config.drop_path_rate = 0.0
        vision_model = AutoModel.from_pretrained(
            model_args.vision_model_name_or_path,
            trust_remote_code=True,
            torch_dtype=torch_dtype,
        )

        # missing_keys, unexpected_keys = model.model.vision_model.load_state_dict(vision_model.state_dict())
        # logger.info(f"missing_keys {missing_keys}")
        # logger.info(f"unexpected_keys {unexpected_keys}")
        # import time
        # time.sleep(torch.distributed.get_rank() * 10)
        # print(vision_model.state_dict())
        # print(model.model.vision_model.state_dict())

        model.model.vision_model = vision_model

    # We resize the embeddings only when necessary to avoid index errors. If you are creating a model from scratch
    # on a small vocab and want a smaller embedding size, remove this test.
    embedding_size = model.get_input_embeddings().weight.shape[0]
    if len(tokenizer) > embedding_size:
        model.resize_token_embeddings(len(tokenizer), pad_to_multiple_of=8)
        new_embedding_size = model.get_input_embeddings().weight.shape[0]
        logger.info(f"{new_embedding_size=}")
    else:
        new_embedding_size = embedding_size

    if model_args.vision_model_freeze:
        model.model.vision_model.requires_grad_(False)
        for name, param in model.named_parameters():
            if ".vision_model." in name:
                param.requires_grad = False
                param.requires_grad_(False)
                logger.info(f"=> set param {name} {param.size()} requires_grad to False.")
            else:
                pass

    if model_args.audio_model_freeze:
        logger.info(f"=> Loading audio model weight...")
        from funasr.train_utils.load_pretrained_model import load_pretrained_model
        load_pretrained_model(
            model=model.model.audio_model.model,
            path="/data/models/FunAudioLLM/SenseVoiceSmall/model.pt",
            ignore_init_mismatch=True,
            oss_bucket=None,
            scope_map=[],
            excludes=None,
        )
        model.model.audio_model.requires_grad_(False)
        for name, param in model.named_parameters():
            if ".audio_model." in name:
                param.requires_grad = False
                param.requires_grad_(False)
                logger.info(f"=> set param {name} {param.size()} requires_grad to False.")
            else:
                pass

    if model_args.language_model_freeze:
        for name, param in model.named_parameters():
            if ".vision_model." in name:
                pass
            elif ".vision_projection." in name:
                pass
            else:
                param.requires_grad = False
                param.requires_grad_(False)
                # logger.info(f"=> set param {name} {param.size()} requires_grad to False.")

        # Unfreeze the new token embeddings
        for idx in range(embedding_size, new_embedding_size):
            model.get_input_embeddings().weight.data[idx].requires_grad = True
            model.get_input_embeddings().weight.data[idx].requires_grad_(True)
            # logger.info(f"=> set input_embedding {idx} requires_grad to True.")

    if getattr(model.config, "model_type", None) == "mixtral" and deepspeed.is_deepspeed_zero3_enabled():
        require_version("deepspeed>=0.13.0", "To fix: pip install deepspeed>=0.13.0")
        from deepspeed.utils import set_z3_leaf_modules
        from transformers.models.mixtral.modeling_mixtral import MixtralSparseMoeBlock
        set_z3_leaf_modules(model, [MixtralSparseMoeBlock])

        model.config.use_cache = False
        if 'output_router_logits' in model.config.to_dict():
            logger.info('set output_router_logits as True')
            model.config.output_router_logits = True
            model.config.router_aux_loss_coef = 0.001

    if training_args.gradient_checkpointing and not model_args.language_model_freeze:
        if hasattr(model, "enable_input_require_grads"):
            model.enable_input_require_grads()
        else:
            def make_inputs_require_grad(module, input, output):
                output.requires_grad = True
                output.requires_grad_(True)
            model.get_input_embeddings().register_forward_hook(make_inputs_require_grad)

    if training_args.gradient_checkpointing:
        if hasattr(model, "vision_model"):
            model.model.vision_model.gradient_checkpointing = True
            model.model.vision_model.encoder.gradient_checkpointing = True
        model.model._set_gradient_checkpointing()

        logger.info(f"Enabling gradient checkpointing")
    
    logger.info(f"model {model}")

    # init mtp layers
    if hasattr(model.config, "num_nextn_predict_layers"):
        for mtp_idx in range(model.config.num_nextn_predict_layers):
            layer_idx = model.config.num_hidden_layers - model.config.num_nextn_predict_layers + mtp_idx

            # ---------------------------------------------------------------------------------------
            # layers
            is_missing = False
            for missing_key in loading_info["missing_keys"]:
                if f"model.layers.{layer_idx}" in missing_key:
                    is_missing = True
                    break

            if is_missing:
                logger.info(f"=> load state dict from layer {layer_idx -1 } to layer {layer_idx}")
                model.model.layers[layer_idx].load_state_dict(model.model.layers[layer_idx - 1].state_dict())
                for param1, param2 in zip(model.model.layers[layer_idx].parameters(), model.model.layers[layer_idx - 1].parameters()):
                    assert torch.equal(param1, param2), "权重复制失败"

            for name, param in model.model.layers[layer_idx].named_parameters():
                param.requires_grad = True
                param.requires_grad_(True)

            # ---------------------------------------------------------------------------------------
            # mtp_projs
            is_missing = False
            for missing_key in loading_info["missing_keys"]:
                if f"mtp_projs.{mtp_idx}" in missing_key:
                    is_missing = True
                    break

            if is_missing and mtp_idx > 0:
                logger.info(f"=> load state dict from mtp_proj {mtp_idx -1 } to mtp_proj {mtp_idx}")
                model.mtp_projs[mtp_idx].load_state_dict(model.mtp_projs[mtp_idx - 1].state_dict())
                for param1, param2 in zip(model.mtp_projs[mtp_idx].parameters(), model.mtp_projs[mtp_idx - 1].parameters()):
                    assert torch.equal(param1, param2), "权重复制失败"

            model.mtp_projs[mtp_idx].requires_grad = True
            model.mtp_projs[mtp_idx].requires_grad_(True)

            # ---------------------------------------------------------------------------------------
            # mtp_embed_norms
            is_missing = False
            for missing_key in loading_info["missing_keys"]:
                if f"mtp_embed_norms.{mtp_idx}" in missing_key:
                    is_missing = True
                    break

            if is_missing and mtp_idx > 0:
                logger.info(f"=> load state dict from mtp_embed_norm {mtp_idx -1 } to mtp_embed_norm {mtp_idx}")
                model.mtp_embed_norms[mtp_idx].load_state_dict(model.mtp_embed_norms[mtp_idx - 1].state_dict())
                for param1, param2 in zip(model.mtp_embed_norms[mtp_idx].parameters(), model.mtp_embed_norms[mtp_idx - 1].parameters()):
                    assert torch.equal(param1, param2), "权重复制失败"

            model.mtp_embed_norms[mtp_idx].requires_grad = True
            model.mtp_embed_norms[mtp_idx].requires_grad_(True)

            # ---------------------------------------------------------------------------------------
            # mtp_hidden_norms
            is_missing = False
            for missing_key in loading_info["missing_keys"]:
                if f"mtp_hidden_norms.{mtp_idx}" in missing_key:
                    is_missing = True
                    break

            if is_missing and mtp_idx > 0:
                logger.info(f"=> load state dict from mtp_hidden_norm {mtp_idx -1 } to mtp_hidden_norm {mtp_idx}")
                model.mtp_hidden_norms[mtp_idx].load_state_dict(model.mtp_hidden_norms[mtp_idx - 1].state_dict())
                for param1, param2 in zip(model.mtp_hidden_norms[mtp_idx].parameters(), model.mtp_hidden_norms[mtp_idx - 1].parameters()):
                    assert torch.equal(param1, param2), "权重复制失败"

            model.mtp_hidden_norms[mtp_idx].requires_grad = True
            model.mtp_hidden_norms[mtp_idx].requires_grad_(True)

    def print_grad_status(model):
        for name, param in model.named_parameters():
            if param.requires_grad:
                logger.info(f"=> parameter {name} requires_grad is True.")

    print_grad_status(model)

    # Load data
    lm_datasets = build_supervised_dataset_deepspeed(
        model_config=config,
        model_args=model_args,
        data_args=data_args,
        training_args=training_args,
        tokenizer=tokenizer,
    )
    tokenized_datasets = lm_datasets

    if training_args.do_train:
        if "train" not in tokenized_datasets:
            raise ValueError("--do_train requires a train dataset")
        train_dataset = lm_datasets["train"]
        if data_args.max_train_samples is not None:
            max_train_samples = min(len(train_dataset), data_args.max_train_samples)
            train_dataset = train_dataset.select(range(max_train_samples))

    if training_args.do_eval:
        if "validation" not in tokenized_datasets:
            raise ValueError("--do_eval requires a validation dataset")
        eval_dataset = lm_datasets["validation"]
        if data_args.max_eval_samples is not None:
            max_eval_samples = min(len(eval_dataset), data_args.max_eval_samples)
            eval_dataset = eval_dataset.select(range(max_eval_samples))

        def preprocess_logits_for_metrics(logits, labels):
            if isinstance(logits, tuple):
                # Depending on the model and config, logits may contain extra tensors,
                # like past_key_values, but logits always come first
                logits = logits[0]
            return logits.argmax(dim=-1)

        metric = evaluate.load("accuracy", cache_dir=model_args.cache_dir)

        def compute_metrics(eval_preds):
            preds, labels = eval_preds
            # preds have the same shape as the labels, after the argmax(-1) has been calculated
            # by preprocess_logits_for_metrics but we need to shift the labels
            labels = labels[:, 1:].reshape(-1)
            preds = preds[:, :-1].reshape(-1)
            return metric.compute(predictions=preds, references=labels)

    if "data_collator" in lm_datasets:
        default_data_collator = lm_datasets["data_collator"]

    # Initialize our Trainer
    trainer = Trainer(
        model=model,
        args=training_args,
        train_dataset=train_dataset if training_args.do_train else None,
        eval_dataset=eval_dataset if training_args.do_eval else None,
        processing_class=tokenizer,
        # Data collator will default to DataCollatorWithPadding, so we change it.
        data_collator=default_data_collator,
        compute_metrics=compute_metrics if training_args.do_eval and not is_torch_xla_available() else None,
        preprocess_logits_for_metrics=preprocess_logits_for_metrics
        if training_args.do_eval and not is_torch_xla_available()
        else None,
    )

    # Training
    if training_args.do_train:
        checkpoint = None
        if training_args.resume_from_checkpoint is not None:
            checkpoint = training_args.resume_from_checkpoint
        elif last_checkpoint is not None:
            checkpoint = last_checkpoint
        train_result = trainer.train(resume_from_checkpoint=checkpoint)
        trainer.save_model()  # Saves the tokenizer too for easy upload

        metrics = train_result.metrics

        max_train_samples = (
            data_args.max_train_samples if data_args.max_train_samples is not None else len(train_dataset)
        )
        metrics["train_samples"] = min(max_train_samples, len(train_dataset))

        trainer.log_metrics("train", metrics)
        trainer.save_metrics("train", metrics)
        trainer.save_state()

    # Evaluation
    if training_args.do_eval:
        logger.info("*** Evaluate ***")

        metrics = trainer.evaluate()

        max_eval_samples = data_args.max_eval_samples if data_args.max_eval_samples is not None else len(eval_dataset)
        metrics["eval_samples"] = min(max_eval_samples, len(eval_dataset))
        try:
            perplexity = math.exp(metrics["eval_loss"])
        except OverflowError:
            perplexity = float("inf")
        metrics["perplexity"] = perplexity

        trainer.log_metrics("eval", metrics)
        trainer.save_metrics("eval", metrics)

    kwargs = {"finetuned_from": model_args.model_name_or_path, "tasks": "text-generation"}
    if data_args.dataset_name is not None:
        kwargs["dataset_tags"] = data_args.dataset_name
        if data_args.dataset_config_name is not None:
            kwargs["dataset_args"] = data_args.dataset_config_name
            kwargs["dataset"] = f"{data_args.dataset_name} {data_args.dataset_config_name}"
        else:
            kwargs["dataset"] = data_args.dataset_name

    if training_args.push_to_hub:
        trainer.push_to_hub(**kwargs)
    else:
        trainer.create_model_card(**kwargs)


def _mp_fn(index):
    # For xla_spawn (TPUs)
    main()


if __name__ == "__main__":
    # use_jit_compile = os.getenv('JIT_COMPILE', 'True').lower() in ['true', '1']
    # use_jit_compile = os.getenv('JIT_COMPILE', 'False').lower() in ['true', '1']
    # torch.npu.set_compile_mode(jit_compile=use_jit_compile)

    # option = {"NPU_FUZZY_COMPILE_BLACKLIST": "Tril,LayerNormGrad"}
    # torch.npu.set_option(option)

    main()