alpaca_lora_finetune.py

import os
import sys
from typing import List

import fire
import torch
import transformers
from datasets import load_dataset, DatasetDict
from transformers import Seq2SeqTrainer, TrainerCallback, TrainingArguments, TrainerState, TrainerControl
from transformers.trainer_utils import PREFIX_CHECKPOINT_DIR

"""
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


class SavePeftModelCallback(TrainerCallback):
    def on_save(
        self,
        args: TrainingArguments,
        state: TrainerState,
        control: TrainerControl,
        **kwargs,
    ):
        checkpoint_folder = os.path.join(args.output_dir, f"{PREFIX_CHECKPOINT_DIR}-{state.global_step}")

        kwargs["model"].save_pretrained(checkpoint_folder)

        pytorch_model_path = os.path.join(checkpoint_folder, "pytorch_model.bin")
        torch.save({}, pytorch_model_path)
        return control


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 = 4,
    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
    add_eos_token: bool = False,
    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.
    # debug mode
    debug_mode: bool = False,
    warmup_steps: int = 100,
):
    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"add_eos_token: {add_eos_token}\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"
            f"debug_mode: {debug_mode}\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)}
        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=True,
        torch_dtype=torch.float16,
        device_map=device_map,
    )

    tokenizer = LlamaTokenizer.from_pretrained(base_model)

    tokenizer.pad_token_id = 0  # unk. we want this to be different from the eos token
    tokenizer.padding_side = "left"  # Allow batched inference

    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=add_eos_token)
            user_prompt_len = len(tokenized_user_prompt["input_ids"])

            if add_eos_token:
                user_prompt_len -= 1

            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)
            if not debug_mode
            else DatasetDict({"train": load_dataset(data_path, split="train[:1024]")})
        )

    if resume_from_checkpoint:
        # Check the available weights and load them
        adapter_checkpoint_name = os.path.join(resume_from_checkpoint, "adapter_model.bin")  # lora checkpoint
        if os.path.exists(adapter_checkpoint_name):
            print(f"Restarting from {adapter_checkpoint_name}")
            adapters_weights = torch.load(adapter_checkpoint_name)
            set_peft_model_state_dict(model, adapters_weights)
        else:
            print(f"Checkpoint {adapter_checkpoint_name} not found")

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

    if val_set_size > 0:
        val_set_size = 128 if debug_mode else val_set_size
        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

    eval_steps = 10 if debug_mode else 200
    save_steps = 10 if debug_mode else 200
    trainer = transformers.Trainer(
        model=model,
        train_dataset=train_data,
        eval_dataset=val_data,
        args=transformers.TrainingArguments(
            per_device_train_batch_size=micro_batch_size,
            gradient_accumulation_steps=gradient_accumulation_steps,
            warmup_steps=warmup_steps,
            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=eval_steps if val_set_size > 0 else None,
            save_steps=save_steps,
            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_collator=transformers.DataCollatorForSeq2Seq(
            tokenizer, pad_to_multiple_of=8, return_tensors="pt", padding=True
        ),
        callbacks=[SavePeftModelCallback],
    )
    model.config.use_cache = False

    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)
    model.base_model.save_pretrained(output_dir)
    pytorch_model_path = os.path.join(output_dir, "pytorch_model.bin")
    torch.save({}, pytorch_model_path)

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


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

Question
it seems to me there are two variables to identify the same folder.
both output_dir and resume_from_checkpoint point to the same set of files. no?

SavePeftModelCallback saves the checkpoints using output_dir. and resume_from_checkpoint loads them.
am I missing something? why isn't this one variable, or one string and one boolean?

class SavePeftModelCallback(TrainerCallback):
        checkpoint_folder = os.path.join(args.output_dir, f"{PREFIX_CHECKPOINT_DIR}-{state.global_step}")

and why are there two sets of instructions to save the model? there is nothing functionally different between them. it's just code repetition? meaning that i could set resum from checkpoint to either and any.?

    model.save_pretrained(output_dir)
    model.base_model.save_pretrained(output_dir)
    torch.save({}, os.path.join(output_dir, "pytorch_model.bin"))

and

    checkpoint_folder = output_dir + state
    kwargs["model"].save_pretrained(checkpoint_folder)
    torch.save({}, os.path.join(checkpoint_folder, "pytorch_model.bin"))

any idea why I am getting

_IncompatibleKeys(missing_keys=['base_model.model.transformer.wte.weight', 'base_model.model.transformer.wpe.weight', 'base_model.model.transformer.h.0.ln_1.weight', 'base_model.model.transformer.h.0.ln_1.bias', 'base_model.model.transformer.h.0.attn.c_attn.weight', 'base_model.model.transformer.h.0.attn.c_attn.bias',
...

?