train.py

# coding=utf-8
# Copyright 2024 Sourab Mangrulkar. 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.

"""
Continued pre-training/fine-tuning of code LLMs for code autocompletion.
"""

import gc
import os
import random
import sys
from typing import Optional
from dataclasses import dataclass, field

import numpy as np
import torch
from datasets import load_dataset
from torch.utils.data import IterableDataset
from tqdm import tqdm
from transformers import (
    AutoModelForCausalLM,
    AutoTokenizer,
    Trainer,
    TrainingArguments,
    HfArgumentParser,
    set_seed,
    BitsAndBytesConfig,
)

from peft import LoraConfig, get_peft_model, prepare_model_for_kbit_training, replace_lora_weights_loftq
import fim


# Define and parse arguments.
@dataclass
class ModelArguments:
    """
    Arguments pertaining to which model/config/tokenizer we are going to fine-tune from.
    """

    model_name_or_path: str = field(
        metadata={
            "help": "Path to pretrained model or model identifier from huggingface.co/models"
        }
    )
    lora_alpha: Optional[int] = field(default=16)
    lora_dropout: Optional[float] = field(default=0.1)
    lora_r: Optional[int] = field(default=64)
    lora_target_modules: Optional[str] = field(
        default="q_proj,k_proj,v_proj,o_proj,down_proj,up_proj,gate_proj",
        metadata={
            "help": "comma separated list of target modules to apply LoRA layers to"
        },
    )
    use_nested_quant: Optional[bool] = field(
        default=False,
        metadata={"help": "Activate nested quantization for 4bit base models"},
    )
    bnb_4bit_compute_dtype: Optional[str] = field(
        default="float16",
        metadata={"help": "Compute dtype for 4bit base models"},
    )
    bnb_4bit_quant_type: Optional[str] = field(
        default="nf4",
        metadata={"help": "Quantization type fp4 or nf4"},
    )
    use_flash_attn: Optional[bool] = field(
        default=False,
        metadata={"help": "Enables Flash attention for training."},
    )
    use_peft_lora: Optional[bool] = field(
        default=False,
        metadata={"help": "Enables PEFT LoRA for training."},
    )
    use_8bit_qunatization: Optional[bool] = field(
        default=False,
        metadata={"help": "Enables loading model in 8bit."},
    )
    use_4bit_quantization: Optional[bool] = field(
        default=False,
        metadata={"help": "Enables loading model in 4bit."},
    )
    use_reentrant: Optional[bool] = field(
        default=False,
        metadata={"help": "Gradient Checkpointing param. Refer the related docs"},
    )
    use_unsloth: Optional[bool] = field(
        default=False,
        metadata={"help": "Enables UnSloth for training."},
    )
    use_loftq: Optional[bool] = field(
        default=False,
        metadata={"help": "Enables LoftQ init for the LoRA adapters when using QLoRA."},
    )
    use_loftq_callback: Optional[bool] = field(
        default=False,
        metadata={"help": "Enables LoftQ callback comparing logits of base model to the ones from LoftQ init. Provides better init."},
    )


@dataclass
class DataTrainingArguments:
    dataset_name: Optional[str] = field(
        default="smangrul/hug_stack",
        metadata={"help": "The preference dataset to use."},
    )
    dataset_text_field: str = field(
        default="text", metadata={"help": "Dataset field to use as input text."}
    )
    max_seq_length: Optional[int] = field(default=4096)
    test_size: Optional[float] = field(default=0.1)
    fim_rate: Optional[float] = field(default=0.5)
    fim_spm_rate: Optional[float] = field(default=0.5)
    splits: Optional[str] = field(
        default="train",
        metadata={"help": "Comma separate list of the splits to use from the dataset."},
    )


def chars_token_ratio(dataset, tokenizer, data_column, nb_examples=400):
    """
    Estimate the average number of characters per token in the dataset.
    """
    total_characters, total_tokens = 0, 0
    for _, example in tqdm(zip(range(nb_examples), iter(dataset)), total=nb_examples):
        total_characters += len(example[data_column])
        total_tokens += len(tokenizer(example[data_column]).tokens())

    return total_characters / total_tokens


class ConstantLengthDataset(IterableDataset):
    """
    Iterable dataset that returns constant length chunks of tokens from stream of text files.
        Args:
            tokenizer (Tokenizer): The processor used for proccessing the data.
            dataset (dataset.Dataset): Dataset with text files.
            infinite (bool): If True the iterator is reset after dataset reaches end else stops.
            seq_length (int): Length of token sequences to return.
            num_of_sequences (int): Number of token sequences to keep in buffer.
            chars_per_token (int): Number of characters per token used to estimate number of tokens in text buffer.
            fim_rate (float): Rate (0.0 to 1.0) that sample will be permuted with FIM.
            fim_spm_rate (float): Rate (0.0 to 1.0) of FIM permuations that will use SPM.
            seed (int): Seed for random number generator.
    """

    def __init__(
        self,
        tokenizer,
        dataset,
        infinite=False,
        seq_length=1024,
        num_of_sequences=1024,
        chars_per_token=3.6,
        content_field="content",
        fim_rate=0.5,
        fim_spm_rate=0.5,
        seed=0,
        shuffle=False,
    ):
        self.tokenizer = tokenizer
        self.concat_token_id = tokenizer.eos_token_id
        self.dataset = dataset
        self.seq_length = seq_length
        self.infinite = infinite
        self.current_size = 0
        self.max_buffer_size = seq_length * chars_per_token * num_of_sequences
        self.content_field = content_field
        self.fim_rate = fim_rate
        self.fim_spm_rate = fim_spm_rate
        self.seed = seed
        self.shuffle = shuffle

        (
            self.bos_token_id,
            self.suffix_tok_id,
            self.prefix_tok_id,
            self.middle_tok_id,
            self.pad_tok_id,
        ) = fim.get_fim_token_ids(self.tokenizer)
        if not self.suffix_tok_id and self.fim_rate > 0:
            print("FIM is not supported by tokenizer, disabling FIM")
            self.fim_rate = 0

    def __iter__(self):
        iterator = iter(self.dataset)
        more_examples = True
        np_rng = np.random.RandomState(seed=self.seed)
        while more_examples:
            buffer, buffer_len = [], 0
            while True:
                if buffer_len >= self.max_buffer_size:
                    break
                try:
                    buffer.append(next(iterator)[self.content_field])
                    buffer_len += len(buffer[-1])
                except StopIteration:
                    if self.infinite:
                        iterator = iter(self.dataset)
                    else:
                        more_examples = False
                        break
            tokenized_inputs = self.tokenizer(
                buffer, truncation=False, add_special_tokens=False
            )["input_ids"]
            all_token_ids = []

            for tokenized_input in tokenized_inputs:
                # optionally do FIM permutations
                if self.fim_rate > 0:
                    tokenized_input, np_rng = fim.permute(
                        tokenized_input,
                        np_rng,
                        self.suffix_tok_id,
                        self.prefix_tok_id,
                        self.middle_tok_id,
                        self.pad_tok_id,
                        fim_rate=self.fim_rate,
                        fim_spm_rate=self.fim_spm_rate,
                        truncate_or_pad=False,
                        bos_token_id=self.bos_token_id,
                    )

                all_token_ids.extend(tokenized_input + [self.concat_token_id])
            examples = []
            for i in range(0, len(all_token_ids), self.seq_length):
                input_ids = all_token_ids[i : i + self.seq_length]
                if len(input_ids) == self.seq_length:
                    examples.append(input_ids)
            if self.shuffle:
                random.shuffle(examples)
            for example in examples:
                self.current_size += 1
                yield {
                    "input_ids": torch.LongTensor(example),
                    "labels": torch.LongTensor(example),
                }


def create_datasets(tokenizer, args, seed):
    dataset = load_dataset(args.dataset_name, split=args.splits)
    dataset = dataset.train_test_split(
        test_size=args.test_size, seed=seed, shuffle=True
    )
    train_data = dataset["train"]
    valid_data = dataset["test"]
    print(
        f"Size of the train set: {len(train_data)}. Size of the validation set: {len(valid_data)}"
    )
    chars_per_token = chars_token_ratio(train_data, tokenizer, args.dataset_text_field)
    print(f"The character to token ratio of the dataset is: {chars_per_token:.2f}")
    train_dataset = ConstantLengthDataset(
        tokenizer,
        train_data,
        infinite=True,
        seq_length=args.max_seq_length,
        chars_per_token=chars_per_token,
        content_field=args.dataset_text_field,
        fim_rate=args.fim_rate,
        fim_spm_rate=args.fim_spm_rate,
        seed=seed,
        shuffle=True,
    )
    valid_dataset = ConstantLengthDataset(
        tokenizer,
        valid_data,
        infinite=False,
        seq_length=args.max_seq_length,
        chars_per_token=chars_per_token,
        content_field=args.dataset_text_field,
        fim_rate=args.fim_rate,
        fim_spm_rate=args.fim_spm_rate,
        seed=seed,
    )
    print(f"A sample of valid dataset: {next(iter(valid_dataset))}")
    return train_dataset, valid_dataset

def get_mae(x, y):
    return (x - y).abs().mean()


def get_mse(x, y):
    return torch.pow(x - y, 2).mean()


def error_report(x, y):
    mae = get_mae(x, y)
    mse = get_mse(x, y)
    print(
        f"Mean absolute error: {mae:>8.5f}\n"
        f"Mean squared error:  {mse:>8.5f}"
    )

    
def loftq_init(model, tokenizer, train_dataset, max_seq_length, args):
    if args.use_loftq_callback:
        compute_dtype = getattr(torch, args.bnb_4bit_compute_dtype)
        base_model = AutoModelForCausalLM.from_pretrained(args.model_name_or_path, torch_dtype=compute_dtype)
        base_model.resize_token_embeddings(len(tokenizer), pad_to_multiple_of=8)
        random_input_ids = torch.randint(0, len(train_dataset), size=(1,)).numpy().tolist()
        random_inputs = [train_dataset[i]['content'] for i in random_input_ids]
        random_inputs = tokenizer(random_inputs, return_tensors="pt", padding=True, truncation="max_length", max_length=max_seq_length)
        logits_base = base_model(**random_inputs).logits
        del base_model
        gc.collect()
        
        def loftq_callback(model, module_name):
            """Callable to replace weights with LoFTQ if the mse is lower than the current best one."""
            global current_mse
            logits = model(**random_inputs).logits
            mse = get_mse(logits_base, logits)
            if mse < current_mse:
                current_mse = mse
                print(f"MSE improved for module {module_name}")
                return True
            print(f"MSE did not improve for module {module_name}")
            return False
        
        replace_lora_weights_loftq(model, callback=loftq_callback)
        logits_loftq_callback = model(**random_inputs).logits
        error_report(logits_base, logits_loftq_callback)
    else:
        replace_lora_weights_loftq(model)


def create_and_prepare_model(args, data_args, training_args):
    device_map = None
    bnb_config = None

    load_in_8bit = args.use_8bit_qunatization
    load_in_4bit = args.use_4bit_quantization

    if args.use_unsloth:
        from unsloth import FastLanguageModel

    if args.use_4bit_quantization:
        compute_dtype = getattr(torch, args.bnb_4bit_compute_dtype)

        bnb_config = BitsAndBytesConfig(
            load_in_4bit=args.use_4bit_quantization,
            bnb_4bit_quant_type=args.bnb_4bit_quant_type,
            bnb_4bit_compute_dtype=compute_dtype,
            bnb_4bit_use_double_quant=args.use_nested_quant,
        )

        if compute_dtype == torch.float16 and args.use_4bit_quantization:
            major, _ = torch.cuda.get_device_capability()
            if major >= 8:
                print("=" * 80)
                print(
                    "Your GPU supports bfloat16, you can accelerate training with the argument --bf16"
                )
                print("=" * 80)

    if args.use_4bit_quantization or args.use_8bit_qunatization:
        device_map = (
            int(os.environ.get("LOCAL_RANK", -1))
            if torch.distributed.is_available() and torch.distributed.is_initialized()
            else "auto"
        )  # {"": 0}

    if args.use_unsloth:
        # Load model
        model, _ = FastLanguageModel.from_pretrained(
            model_name=args.model_name_or_path,
            max_seq_length=data_args.max_seq_length,
            dtype=None,
            load_in_4bit=load_in_4bit,
        )
    else:
        model = AutoModelForCausalLM.from_pretrained(
            args.model_name_or_path,
            load_in_8bit=load_in_8bit,
            quantization_config=bnb_config,
            device_map=device_map,
            trust_remote_code=True,
            attn_implementation="flash_attention_2" if args.use_flash_attn else "eager",
        )

    if (
        (args.use_4bit_quantization or args.use_8bit_qunatization)
        and args.use_peft_lora
        and not args.use_unsloth
    ):
        model = prepare_model_for_kbit_training(
            model,
            use_gradient_checkpointing=training_args.gradient_checkpointing,
            gradient_checkpointing_kwargs={"use_reentrant": model_args.use_reentrant},
        )

    if args.use_peft_lora and not args.use_unsloth:
        peft_config = LoraConfig(
            lora_alpha=args.lora_alpha,
            lora_dropout=args.lora_dropout,
            r=args.lora_r,
            bias="none",
            task_type="CAUSAL_LM",
            target_modules=args.lora_target_modules.split(",")
            if args.lora_target_modules != "all-linear"
            else args.lora_target_modules,
        )
        model = get_peft_model(model, peft_config)
    elif args.use_peft_lora and args.use_unsloth:
        # Do model patching and add fast LoRA weights
        model = FastLanguageModel.get_peft_model(
            model,
            lora_alpha=args.lora_alpha,
            lora_dropout=args.lora_dropout,
            r=args.lora_r,
            target_modules=args.lora_target_modules.split(",")
            if args.lora_target_modules != "all-linear"
            else args.lora_target_modules,
            use_gradient_checkpointing=training_args.gradient_checkpointing,
            random_state=training_args.seed,
            max_seq_length=data_args.max_seq_length,
        )
    return model


def main(model_args, data_args, training_args):
    # Set seed for reproducibility
    set_seed(training_args.seed)

    # load the tokenizer
    tokenizer = AutoTokenizer.from_pretrained(model_args.model_name_or_path)

    # load the datasets
    train_dataset, eval_dataset = create_datasets(
        tokenizer, data_args, training_args.seed
    )
    train_dataset.start_iteration = 0

    model = create_and_prepare_model(model_args, data_args, training_args)
    # gradient ckpt
    model.config.use_cache = not training_args.gradient_checkpointing
    training_args.gradient_checkpointing = (
        training_args.gradient_checkpointing and not model_args.use_unsloth
    )
    if training_args.gradient_checkpointing:
        training_args.gradient_checkpointing_kwargs = {
            "use_reentrant": model_args.use_reentrant
        }

    # trainer
    trainer = Trainer(
        model=model,
        args=training_args,
        train_dataset=train_dataset,
        eval_dataset=eval_dataset,
    )
    trainer.accelerator.print(f"{trainer.model}")
    if model_args.use_peft_lora:
        trainer.model.print_trainable_parameters()

    # LoftQ initialization when using QLoRA
    if model_args.use_4bit_quantization and model_args.use_loftq:
        loftq_init(trainer.model, tokenizer, train_dataset, data_args.max_seq_length ,model_args)

    # train
    checkpoint = None
    if training_args.resume_from_checkpoint is not None:
        checkpoint = training_args.resume_from_checkpoint
    trainer.train(resume_from_checkpoint=checkpoint)

    # saving final model
    if trainer.is_fsdp_enabled:
        trainer.accelerator.state.fsdp_plugin.set_state_dict_type("FULL_STATE_DICT")
    trainer.save_model()


if __name__ == "__main__":
    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()
    main(model_args, data_args, training_args)