finetune_fastchat.py

import os
import sys
from typing import List

import fire
import torch
import transformers
from datasets import load_dataset

"""
Unused imports:
import torch.nn as nn
import bitsandbytes as bnb
"""

from peft import (
    LoraConfig,
    get_peft_model,
    get_peft_model_state_dict,
    prepare_model_for_int8_training,
    set_peft_model_state_dict,
)
from transformers import LlamaForCausalLM, LlamaTokenizer

from utils.prompter import Prompter
import copy
from dataclasses import dataclass, field
import json
import logging
import pathlib
from typing import Dict, Optional, Sequence
from torch.utils.data import Dataset

from fastchat import conversation as conversation_lib


IGNORE_INDEX = -100
DEFAULT_PAD_TOKEN = "[PAD]"
DEFAULT_EOS_TOKEN = "</s>"
DEFAULT_BOS_TOKEN = "</s>"
DEFAULT_UNK_TOKEN = "</s>"


def smart_tokenizer_and_embedding_resize(
    special_tokens_dict: Dict,
    tokenizer: transformers.PreTrainedTokenizer,
    model: transformers.PreTrainedModel,
):
    """Resize tokenizer and embedding.

    Note: This is the unoptimized version that may make your embedding size not be divisible by 64.
    """
    num_new_tokens = tokenizer.add_special_tokens(special_tokens_dict)
    model.resize_token_embeddings(len(tokenizer))

    if num_new_tokens > 0:
        input_embeddings = model.get_input_embeddings().weight.data
        output_embeddings = model.get_output_embeddings().weight.data

        input_embeddings_avg = input_embeddings[:-num_new_tokens].mean(
            dim=0, keepdim=True)
        output_embeddings_avg = output_embeddings[:-num_new_tokens].mean(
            dim=0, keepdim=True)

        input_embeddings[-num_new_tokens:] = input_embeddings_avg
        output_embeddings[-num_new_tokens:] = output_embeddings_avg


def _tokenize_fn(strings: Sequence[str],
                 tokenizer: transformers.PreTrainedTokenizer) -> Dict:
    """Tokenize a list of strings."""
    tokenized_list = [
        tokenizer(
            text,
            return_tensors="pt",
            padding="longest",
            max_length=tokenizer.model_max_length,
            truncation=True,
        ) for text in strings
    ]
    input_ids = labels = [
        tokenized.input_ids[0] for tokenized in tokenized_list
    ]
    input_ids_lens = labels_lens = [
        tokenized.input_ids.ne(tokenizer.pad_token_id).sum().item()
        for tokenized in tokenized_list
    ]
    return dict(
        input_ids=input_ids,
        labels=labels,
        input_ids_lens=input_ids_lens,
        labels_lens=labels_lens,
    )


def _mask_targets(target, tokenized_lens, speakers, header_len, s_ids):
    cur_idx = header_len
    tgt_len = target.shape[0]
    for tokenized_len, speaker, s_id in zip(tokenized_lens, speakers, s_ids):
        if cur_idx >= tgt_len:
            break
        elif cur_idx + tokenized_len < tgt_len:
            # Check whether the mask is applied to the correct position
            if not torch.equal(target[cur_idx + 2:cur_idx + tokenized_len],
                               s_id[2:]):
                logging.warning("a sentence mismatches the corresponding piece "
                                "in the conversation")
        if speaker == "human":
            target[cur_idx:cur_idx + tokenized_len] = IGNORE_INDEX
        cur_idx += tokenized_len


def _add_speaker_and_signal(header, source, get_conversation=True):
    """Add speaker and start/end signal on each round."""
    BEGIN_SIGNAL = "### "
    END_SIGNAL = "\n"
    conversation = header
    unknown_role = "unknown"  # use default unknown role
    roles = {
        "human": conversation_lib.default_conversation.roles[0],  # human role
        "gpt": conversation_lib.default_conversation.roles[1],  # gpt role
    }
    for sentence in source:
        sentence_from = sentence["from"].lower()
        sentence["value"] = (
            BEGIN_SIGNAL
            + roles.get(sentence_from, unknown_role)
            + ": "
            + sentence["value"]
            + END_SIGNAL
        )
        if get_conversation:
            conversation += sentence["value"]
    return conversation


def preprocess(
    sources: Sequence[str],
    tokenizer: transformers.PreTrainedTokenizer,
) -> Dict:
    """
    Given a list of sources, each is a conversation list. This transform:
    1. Add signal '### ' at the beginning each sentence, with end signal '\n';
    2. Concatenate conversations together;
    3. Tokenize the concatenated conversation;
    4. Make a deepcopy as the target. Mask human words with IGNORE_INDEX.
    """
    # add end signal and concatenate together
    conversations = []
    header = f"{conversation_lib.default_conversation.system}\n\n"
    for source in sources:
        conversation = _add_speaker_and_signal(header, source)
        conversations.append(conversation)
    # tokenize conversations
    conversations_tokenized = _tokenize_fn(conversations, tokenizer)
    input_ids = conversations_tokenized["input_ids"]
    targets = copy.deepcopy(input_ids)
    header_len = _tokenize_fn([header], tokenizer)["input_ids_lens"][0]
    for target, source in zip(targets, sources):
        tokenized_sentence = _tokenize_fn([s["value"] for s in source], tokenizer)
        tokenized_lens = tokenized_sentence["input_ids_lens"]
        # Currently, "###" is tokenized into 2 tokens in the whole conversation,
        # and 1 token in a single sentence, so we do not need to use the line below.
        # tokenized_lens = [l-1 for l in tokenized_lens]
        speakers = [sentence["from"] for sentence in source]
        ids = tokenized_sentence["input_ids"]
        _mask_targets(target, tokenized_lens, speakers, header_len, ids)

    return dict(input_ids=input_ids, labels=targets)


class SupervisedDataset(Dataset):
    """Dataset for supervised fine-tuning."""

    def __init__(self, data_path: str,
                 tokenizer: transformers.PreTrainedTokenizer):
        super(SupervisedDataset, self).__init__()
        logging.warning("Loading data...")
        list_data_dict = json.load(open(data_path, "r"))

        logging.warning("Formatting inputs...")
        sources = [example["conversations"] for example in list_data_dict]
        data_dict = preprocess(sources, tokenizer)

        self.input_ids = data_dict["input_ids"]
        self.labels = data_dict["labels"]

    def __len__(self):
        return len(self.input_ids)

    def __getitem__(self, i) -> Dict[str, torch.Tensor]:
        return dict(input_ids=self.input_ids[i], labels=self.labels[i])


class LazySupervisedDataset(Dataset):
    """Dataset for supervised fine-tuning."""

    def __init__(self, data_path: str,
                 tokenizer: transformers.PreTrainedTokenizer):
        super(LazySupervisedDataset, self).__init__()
        logging.warning("Loading data...")
        list_data_dict = json.load(open(data_path, "r"))

        logging.warning("Formatting inputs...Skip in lazy mode")
        self.tokenizer = tokenizer
        self.list_data_dict = list_data_dict

    def __len__(self):
        return len(self.list_data_dict)

    def __getitem__(self, i) -> Dict[str, torch.Tensor]:
        sources = self.list_data_dict[i]
        if isinstance(i, int):
            sources = [sources]
        data_dict = preprocess(
            copy.deepcopy([e["conversations"] for e in sources]),
            self.tokenizer)
        if isinstance(i, int):
            data_dict = dict(input_ids=data_dict["input_ids"][0],
                             labels=data_dict["labels"][0])
        return data_dict


@dataclass
class DataCollatorForSupervisedDataset(object):
    """Collate examples for supervised fine-tuning."""

    tokenizer: transformers.PreTrainedTokenizer

    def __call__(self, instances: Sequence[Dict]) -> Dict[str, torch.Tensor]:
        input_ids, labels = tuple([instance[key] for instance in instances]
                                  for key in ("input_ids", "labels"))
        input_ids = torch.nn.utils.rnn.pad_sequence(
            input_ids,
            batch_first=True,
            padding_value=self.tokenizer.pad_token_id)
        labels = torch.nn.utils.rnn.pad_sequence(labels,
                                                 batch_first=True,
                                                 padding_value=IGNORE_INDEX)
        return dict(
            input_ids=input_ids,
            labels=labels,
            attention_mask=input_ids.ne(self.tokenizer.pad_token_id),
        )


def make_supervised_data_module(data_path,tokenizer: transformers.PreTrainedTokenizer,
                                lazy_preprocess=True) -> Dict:
    """Make dataset and collator for supervised fine-tuning."""
    dataset_cls = (LazySupervisedDataset
                   if lazy_preprocess else SupervisedDataset)
    train_dataset = dataset_cls(tokenizer=tokenizer,
                                data_path=data_path)
    data_collator = DataCollatorForSupervisedDataset(tokenizer=tokenizer)
    return dict(train_dataset=train_dataset,
                eval_dataset=None,
                data_collator=data_collator)


def train(
    # model/data params
    base_model: str = "",  # the only required argument
    data_path: str = "yahma/alpaca-cleaned",
    output_dir: str = "./lora-alpaca",
    # training hyperparams
    batch_size: int = 128,
    micro_batch_size: int = 2,
    num_epochs: int = 3,
    learning_rate: float = 3e-4,
    cutoff_len: int = 256,
    val_set_size: int = 2000,
    # lora hyperparams
    lora_r: int = 8,
    lora_alpha: int = 16,
    lora_dropout: float = 0.05,
    lora_target_modules: List[str] = [
        "q_proj",
        "v_proj",
    ],
    # llm hyperparams
    train_on_inputs: bool = True,  # if False, masks out inputs in loss
    group_by_length: bool = False,  # faster, but produces an odd training loss curve
    # wandb params
    wandb_project: str = "",
    wandb_run_name: str = "",
    wandb_watch: str = "",  # options: false | gradients | all
    wandb_log_model: str = "",  # options: false | true
    resume_from_checkpoint: str = None,  # either training checkpoint or final adapter
    prompt_template_name: str = "alpaca",  # The prompt template to use, will default to alpaca.
):
    if int(os.environ.get("LOCAL_RANK", 0)) == 0:
        print(
            f"Training Alpaca-LoRA model with params:\n"
            f"base_model: {base_model}\n"
            f"data_path: {data_path}\n"
            f"output_dir: {output_dir}\n"
            f"batch_size: {batch_size}\n"
            f"micro_batch_size: {micro_batch_size}\n"
            f"num_epochs: {num_epochs}\n"
            f"learning_rate: {learning_rate}\n"
            f"cutoff_len: {cutoff_len}\n"
            f"val_set_size: {val_set_size}\n"
            f"lora_r: {lora_r}\n"
            f"lora_alpha: {lora_alpha}\n"
            f"lora_dropout: {lora_dropout}\n"
            f"lora_target_modules: {lora_target_modules}\n"
            f"train_on_inputs: {train_on_inputs}\n"
            f"group_by_length: {group_by_length}\n"
            f"wandb_project: {wandb_project}\n"
            f"wandb_run_name: {wandb_run_name}\n"
            f"wandb_watch: {wandb_watch}\n"
            f"wandb_log_model: {wandb_log_model}\n"
            f"resume_from_checkpoint: {resume_from_checkpoint or False}\n"
            f"prompt template: {prompt_template_name}\n"
        )
    assert (
        base_model
    ), "Please specify a --base_model, e.g. --base_model='huggyllama/llama-7b'"
    gradient_accumulation_steps = batch_size // micro_batch_size

    prompter = Prompter(prompt_template_name)

    device_map = "auto"
    world_size = int(os.environ.get("WORLD_SIZE", 1))
    ddp = world_size != 1
    if ddp:
        #device_map = {"": int(os.environ.get("LOCAL_RANK") or 0)}
        device_map = "auto"
        rank = int(os.environ.get("LOCAL_RANK", 0))
        print("ddp,rank:",rank)
        """
        if rank == 0:
            os.environ["CUDA_VISIBLE_DEVICES"]="0,2"
        elif rank == 1:
            os.environ["CUDA_VISIBLE_DEVICES"]="1,3"
        else:
            raise
        """
        gradient_accumulation_steps = gradient_accumulation_steps // world_size

    # Check if parameter passed or if set within environ
    use_wandb = len(wandb_project) > 0 or (
        "WANDB_PROJECT" in os.environ and len(os.environ["WANDB_PROJECT"]) > 0
    )
    # Only overwrite environ if wandb param passed
    if len(wandb_project) > 0:
        os.environ["WANDB_PROJECT"] = wandb_project
    if len(wandb_watch) > 0:
        os.environ["WANDB_WATCH"] = wandb_watch
    if len(wandb_log_model) > 0:
        os.environ["WANDB_LOG_MODEL"] = wandb_log_model

    model = LlamaForCausalLM.from_pretrained(
        base_model,
        load_in_8bit=False,
        torch_dtype=torch.float16,
        device_map=device_map,
    )

    tokenizer = LlamaTokenizer.from_pretrained(
        base_model,
        model_max_length=1024,
        padding_side="right",
        use_fast=False
    )

    if tokenizer.pad_token is None:
        smart_tokenizer_and_embedding_resize(
            special_tokens_dict=dict(pad_token=DEFAULT_PAD_TOKEN),
            tokenizer=tokenizer,
            model=model,
        )
    tokenizer.add_special_tokens({
            "eos_token": DEFAULT_EOS_TOKEN,
            "bos_token": DEFAULT_BOS_TOKEN,
            "unk_token": DEFAULT_UNK_TOKEN,
        })

    def tokenize(prompt, add_eos_token=True):
        # there's probably a way to do this with the tokenizer settings
        # but again, gotta move fast
        result = tokenizer(
            prompt,
            truncation=True,
            max_length=cutoff_len,
            padding=False,
            return_tensors=None,
        )
        if (
            result["input_ids"][-1] != tokenizer.eos_token_id
            and len(result["input_ids"]) < cutoff_len
            and add_eos_token
        ):
            result["input_ids"].append(tokenizer.eos_token_id)
            result["attention_mask"].append(1)

        result["labels"] = result["input_ids"].copy()

        return result

    def generate_and_tokenize_prompt(data_point):
        full_prompt = prompter.generate_prompt(
            data_point["instruction"],
            data_point["input"],
            data_point["output"],
        )
        tokenized_full_prompt = tokenize(full_prompt)
        if not train_on_inputs:
            user_prompt = prompter.generate_prompt(
                data_point["instruction"], data_point["input"]
            )
            tokenized_user_prompt = tokenize(user_prompt, add_eos_token=False)
            user_prompt_len = len(tokenized_user_prompt["input_ids"])

            tokenized_full_prompt["labels"] = [
                -100
            ] * user_prompt_len + tokenized_full_prompt["labels"][
                user_prompt_len:
            ]  # could be sped up, probably
        return tokenized_full_prompt

    #model = prepare_model_for_int8_training(model)

    config = LoraConfig(
        r=lora_r,
        lora_alpha=lora_alpha,
        target_modules=lora_target_modules,
        lora_dropout=lora_dropout,
        bias="none",
        task_type="CAUSAL_LM",
    )
    model = get_peft_model(model, config)

    #if data_path.endswith(".json") or data_path.endswith(".jsonl"):
    #    data = load_dataset("json", data_files=data_path)
    #else:
    #    data = load_dataset(data_path)
        
    data_module = make_supervised_data_module(data_path,tokenizer=tokenizer)
    

    if resume_from_checkpoint:
        # Check the available weights and load them
        checkpoint_name = os.path.join(
            resume_from_checkpoint, "pytorch_model.bin"
        )  # Full checkpoint
        if not os.path.exists(checkpoint_name):
            checkpoint_name = os.path.join(
                resume_from_checkpoint, "adapter_model.bin"
            )  # only LoRA model - LoRA config above has to fit
            resume_from_checkpoint = (
                False  # So the trainer won't try loading its state
            )
        # The two files above have a different name depending on how they were saved, but are actually the same.
        if os.path.exists(checkpoint_name):
            print(f"Restarting from {checkpoint_name}")
            adapters_weights = torch.load(checkpoint_name)
            model = set_peft_model_state_dict(model, adapters_weights)
        else:
            print(f"Checkpoint {checkpoint_name} not found")

    model.print_trainable_parameters()  # Be more transparent about the % of trainable params.

    val_set_size = 0
    """
    if val_set_size > 0:
        train_val = data["train"].train_test_split(
            test_size=val_set_size, shuffle=True, seed=42
        )
        train_data = (
            train_val["train"].shuffle().map(generate_and_tokenize_prompt)
        )
        val_data = (
            train_val["test"].shuffle().map(generate_and_tokenize_prompt)
        )
    else:
        train_data = data["train"].shuffle().map(generate_and_tokenize_prompt)
        val_data = None
    """

    if not ddp and torch.cuda.device_count() > 1:
        # keeps Trainer from trying its own DataParallelism when more than 1 gpu is available
        model.is_parallelizable = True
        model.model_parallel = True

    trainer = transformers.Trainer(
        model=model,
        
        args=transformers.TrainingArguments(
            per_device_train_batch_size=micro_batch_size,
            gradient_accumulation_steps=gradient_accumulation_steps,
            warmup_steps=100,
            num_train_epochs=num_epochs,
            learning_rate=learning_rate,
            fp16=True,
            logging_steps=10,
            optim="adamw_torch",
            evaluation_strategy="steps" if val_set_size > 0 else "no",
            save_strategy="steps",
            eval_steps=200 if val_set_size > 0 else None,
            save_steps=200,
            output_dir=output_dir,
            save_total_limit=3,
            load_best_model_at_end=True if val_set_size > 0 else False,
            ddp_find_unused_parameters=False if ddp else None,
            group_by_length=group_by_length,
            report_to="wandb" if use_wandb else None,
            run_name=wandb_run_name if use_wandb else None,
        ),
        **data_module,
    )
    model.config.use_cache = False

    old_state_dict = model.state_dict
    model.state_dict = (
        lambda self, *_, **__: get_peft_model_state_dict(
            self, old_state_dict()
        )
    ).__get__(model, type(model))

    if torch.__version__ >= "2" and sys.platform != "win32":
        model = torch.compile(model)

    trainer.train(resume_from_checkpoint=resume_from_checkpoint)

    model.save_pretrained(output_dir)

    print(
        "\n If there's a warning about missing keys above, please disregard :)"
    )


if __name__ == "__main__":
    fire.Fire(train)