03_train_1.4B_minipile.py

# Adapted from 02_eval_160M.ipynb
# Training script for Distributed Training of Pythia 1.4B on MiniPile

import os
import torch
import numpy as np
from tqdm import tqdm
from pathlib import Path
from torch.optim import Adam
from datasets import load_dataset
from huggingface_hub import snapshot_download
from torch.optim.lr_scheduler import _LRScheduler
from transformers import DataCollatorForLanguageModeling
from transformers import AutoModelForCausalLM, AutoTokenizer, Trainer, TrainingArguments, get_cosine_schedule_with_warmup

base_dir = "/vol/tmp/koppelmm"
base_path = Path(base_dir)

class CosineSchedulerWithMinLR(_LRScheduler):
    # Basically wrapping the get_cosing_schedule_with_warmup in a lower-bound setting
    # Allows for Cosine Scheduling with a min_lr enforced
    def __init__(self, optimizer, num_warmup_steps, num_training_steps, min_lr):
        self.min_lr = min_lr
        self.base_scheduler = get_cosine_schedule_with_warmup(
            optimizer=optimizer,
            num_warmup_steps=num_warmup_steps,
            num_training_steps=num_training_steps
        )
        super().__init__(optimizer)
        
    def get_lr(self):
        lrs = self.base_scheduler.get_lr()
        return [max(lr, self.min_lr) for lr in lrs]
        
    def step(self):
        self.base_scheduler.step()
        super().step() 

def download_model(down_dir: str, target_folder: str, cache_folder: str, repo_id: str, branch: str = "main") -> None:
    down_dir = Path(down_dir)
    target_dir = down_dir / target_folder
    cache_dir = down_dir / cache_folder

    os.makedirs(target_dir, exist_ok=True)
    os.makedirs(cache_dir, exist_ok=True)

    print(f"Downloading {repo_id}/{branch}...")

    while True:
        try:
            snapshot_download(
                repo_id,
                repo_type="model",
                revision=branch,
                cache_dir=str(cache_dir),
                local_dir=str(target_dir)
            )
            break
        except Exception as e:
            print(f"Download attempt failed: {e}")
            continue

def training():
    #download_model(down_dir=base_dir, target_folder="pythia1.4b_dedup_untrained", 
    #               cache_folder="pythia1.4b_dedup_untrained_Cache",
    #               repo_id="EleutherAI/pythia-1.4b-deduped", branch="step0")

    #download_model(down_dir=base_dir, target_folder="pythia1.4b_dedup_pile", 
    #               cache_folder="pythia1.4b_dedup_pile_Cache",
    #               repo_id="EleutherAI/pythia-1.4b-deduped", branch="main")

    minipile_train = load_dataset("parquet",
                                  data_files={
                                      "train": str(base_path / "MiniPile" / "data/train-*.parquet"),
                                  },
                                  cache_dir=str(base_path / "MiniPile_Cache"),
                                  split="train")

    minipile_val = load_dataset("parquet",
                                data_files={
                                    "validation": str(base_path / "MiniPile" / "data/validation-*.parquet"),
                                },
                                cache_dir=str(base_path / "MiniPile_Cache"),
                                split="validation")

    tokenizer = AutoTokenizer.from_pretrained(base_path / "pythia1.4b_dedup_untrained", use_fast=True, local_files_only=True)
    empty_model = AutoModelForCausalLM.from_pretrained(base_path / "pythia1.4b_dedup_untrained", local_files_only=True, low_cpu_mem_usage=True)

    # Tokenizer doesn't have a pad token, use EOS as a substitute
    if tokenizer.pad_token is None:
        tokenizer.pad_token = tokenizer.eos_token
        tokenizer.pad_token_id = tokenizer.eos_token_id

    def tokenize(example): 
        # seq_len = max_length = 2048 (as upper boundary, so not strict size -> no padding needed)
        return tokenizer(example["text"], 
                         truncation=True, 
                         max_length=2048,
                         return_special_tokens_mask=True)

    # I deleted old iterations of this folder before running this script for first time
    if os.path.exists(base_path / "minipile_train_tokenized"):
        minipile_train_tokenized = load_dataset("arrow", data_files=str(base_path / "minipile_train_tokenized/*.arrow"), split="train")
        minipile_val_tokenized = load_dataset("arrow", data_files=str(base_path / "minipile_val_tokenized/*.arrow"), split="train")
    else:
        minipile_train_tokenized = minipile_train.map(tokenize, batched=True, remove_columns=minipile_train.column_names) # retain only new fields from tokenization
        minipile_val_tokenized = minipile_val.map(tokenize, batched=True, remove_columns=minipile_val.column_names)
        minipile_train_tokenized.save_to_disk(base_path / "minipile_train_tokenized")
        minipile_val_tokenized.save_to_disk(base_path / "minipile_val_tokenized")

    batch_size = 2 #4 #8 is too much for 3x A100
    total_batch = 1024

    # Dynamic padding during training (mlm -> mask language model -> we're doing causal here)
    # https://huggingface.co/docs/transformers/v4.48.0/en/main_classes/data_collator#transformers.DataCollatorForLanguageModeling
    data_collator = DataCollatorForLanguageModeling(tokenizer=tokenizer, mlm=False, pad_to_multiple_of=8)

    seed = 42
    torch.manual_seed(seed)
    np.random.seed(seed)

    if torch.cuda.is_available():
        torch.cuda.manual_seed_all(seed)
        device_count = torch.cuda.device_count()
        print(f"Available GPUs: {device_count}")
        for i in range(device_count):
            device = torch.device(f'cuda:{i}')
            device_properties = torch.cuda.get_device_properties(device)
            total_mem = device_properties.total_memory / (1024 ** 3)
            allocd_mem = torch.cuda.memory_allocated(device) / (1024 ** 3)
            free_mem = total_mem - allocd_mem
            print(f"\nGPU {i}:\t{device_properties.name}")
            print(f"\tTotal memory:\t\t{total_mem:.2f} GiB")
            print(f"\tAllocated memory:\t{allocd_mem:5.2f} GiB")
            print(f"\tFree memory:\t\t{free_mem:.2f} GiB")
    else:
        print("No CUDA-capable GPUs available")

    output_dir = str(base_path / "pythia1.4b_minipile_trained")
    log_dir = str(base_path / "1.4b_minipile_logs")
    os.makedirs(output_dir, exist_ok=True)
    os.makedirs(log_dir, exist_ok=True)

    # https://huggingface.co/docs/transformers/v4.46.0/en/main_classes/trainer#transformers.TrainingArguments
    training_args = TrainingArguments(
        output_dir=output_dir,
        overwrite_output_dir=True,
        num_train_epochs=1.5,            # Since train_iters gets set, use num_train_epochs=1.5 like for The Pile
        per_device_train_batch_size=batch_size,   # Gives an effective batch size of 1024 after grad accum
        per_device_eval_batch_size=batch_size,    # Same as training batch size
        gradient_accumulation_steps=(total_batch // batch_size // device_count), # Achieve a batch size of 1024
        learning_rate=2e-4,              # Default Pythia 1.4B-specific
        weight_decay=0.01,               # Default Pythia 160M and 1.4B
        max_steps=1024,                  # Adjusted for MiniPile (https://discuss.huggingface.co/t/how-does-max-steps-affect-the-number-of-samples-the-model-sees/69681)
        warmup_steps=int(0.01 * 1024),   # 1% of total steps for warmup
        logging_dir=log_dir,
        logging_steps=10,
        eval_strategy="steps",
        eval_steps=100,     # Frequency for evaluation during training
        save_steps=1024,    # Save at the end of training
        save_total_limit=1, # Only keep the most recent checkpoint
        fp16=True,          # Using mixed precision for comparable conditions
        report_to=None,     # Noting this for later iterations, maybe set this as "wandb", "tensorboard" or smth
        ddp_find_unused_parameters=False, # see https://discuss.pytorch.org/t/how-to-change-ddp-parameter-find-unused-parameters-true-to-false-during-training/130763
        max_grad_norm=1.0,  # Default Pythia 160M and 1.4B
        dataloader_num_workers=4,
        dataloader_pin_memory=True,
        #gradient_checkpointing=True,
    )

    # Ensure training across multiple GPUs if available
    device = "cuda" if torch.cuda.is_available() else "cpu"
    empty_model = empty_model.to(device)

    optimizer = Adam(empty_model.parameters(), lr=training_args.learning_rate, betas=(0.9, 0.95), eps=1e-8, weight_decay=0.01)

    scheduler = CosineSchedulerWithMinLR(
        optimizer=optimizer,
        num_warmup_steps=training_args.warmup_steps,
        num_training_steps=training_args.max_steps,
        min_lr=2e-5 # Pythia 1.4B-specific
    )

    # Train Pythia 1.4B Untrained on MiniPile
    # https://huggingface.co/docs/transformers/v4.46.0/en/main_classes/trainer
    trainer = Trainer(model=empty_model,
                    args=training_args,
                    train_dataset=minipile_train_tokenized,
                    eval_dataset=minipile_val_tokenized,
                    data_collator=data_collator,
                    optimizers=(optimizer, scheduler))

    trainer.train()

    # Why is this a two-step process?!
    trainer.save_model(str(base_path / "pythia1.4b_minipile_trained")) # This saves the model weights

if __name__ == "__main__":
    training()
    torch.cuda.empty_cache()

# tmux new -s 14b_minipile
# conda activate minicorpus
# Detach from tmux session: Ctrl-b followed by d
# Reattach to tmux session: tmux attach -t 14b_minipile
# tmux list-sessions
# tmux kill-session -t 14b_minipile