train.py

"""
This training script can be run both on a single gpu in debug mode,
and also in a larger training run with distributed data parallel (ddp).

To run in debug mode example:
$ python train.py --batch_size=32 --other=args

To run DDP on 4 gpus on one node, example:
$ torchrun --standalone --nproc_per_node=4 train.py
"""

import os
import sys
import time
import math
from ast import literal_eval

import wandb
import numpy as np
import torch
from torch.nn.parallel import DistributedDataParallel as DDP
from torch.distributed import init_process_group, destroy_process_group

from model import GPTConfig, GPT

# -----------------------------------------------------------------------------
# default config values
# I/O
out_dir = 'out'
eval_interval = 500
log_interval = 1
eval_iters = 50
eval_only = False # if True, script exits right after the first eval
always_save_checkpoint = False # if True, always save a checkpoint after each eval
# wandb logging
wandb_log = False # disabled by default
wandb_entity = 'karpathy'
wandb_project = 'owt'
wandb_run_name = 'gpt2' # 'run' + str(time.time())
# data
dataset = 'openwebtext'
batch_size = 8
block_size = 1024
# model
device = 'cuda:0'
init_from = 'scratch' # 'scratch' or 'resume' or 'gpt2*'
dropout = 0.1
n_layer = 12
n_head = 12
n_embd = 768
# adamw optimizer
learning_rate = 2.5e-4 # max learning rate
max_iters = 500000 # total number of training iterations
weight_decay = 1e-2
betas = (0.9, 0.95)
# learning rate decay settings
decay_lr = True # whether to decay the learning rate
warmup_iters = 2000 # how many steps to warm up for
lr_decay_iters = 320000 # how many steps to decay the learning rate for
min_lr = 1e-5 # minimum learning rate
# DDP settings
backend = 'nccl' # 'nccl', 'gloo', etc.
compile = True # use PyTorch 2.0 to compile the model to be faster
# -----------------------------------------------------------------------------
# poor man's Configurator. Potentially a bad idea. Example usage:
# $ python train.py override_file --batch_size=32
# this will first run config/override_file.py, then override batch_size to 32
for arg in sys.argv[1:]:
    if '=' not in arg:
        # assume it's the name of a config file
        assert not arg.startswith('--')
        config_file = os.path.join('config', arg + '.py')
        print(f"Overriding config with {config_file}:")
        with open(config_file) as f:
            print(f.read())
        exec(open(config_file).read())
    else:
        # assume it's a --key=value argument
        assert arg.startswith('--')
        key, val = arg.split('=')
        key = key[2:]
        if key in globals():
            try:
                # attempt to eval it it (e.g. if bool, number, or etc)
                attempt = literal_eval(val)
            except (SyntaxError, ValueError):
                # if that goes wrong, just use the string
                attempt = val
            # ensure the types match ok
            assert type(attempt) == type(globals()[key])
            # cross fingers
            print(f"Overriding: {key} = {attempt}")
            globals()[key] = attempt
        else:
            raise ValueError(f"Unknown config key: {key}")
# -----------------------------------------------------------------------------
ddp = int(os.environ.get('LOCAL_RANK', -1)) != -1 # is this a ddp run?
if ddp:
    init_process_group(backend=backend)
    gpu_id = int(os.environ["LOCAL_RANK"])
    device = f"cuda:{gpu_id}"
else:
    gpu_id = 0 # gpu_id 0 means this is the (single) master process, basically

if gpu_id == 0:
    os.makedirs(out_dir, exist_ok=True)
torch.manual_seed(1337 + gpu_id) # note: each worker gets a different seed
torch.backends.cuda.matmul.allow_tf32 = True # allow tf32 on matmul
torch.backends.cudnn.allow_tf32 = True # allow tf32 on cudnn

# poor man's data loader, TODO evaluate need for actual DataLoader
data_dir = os.path.join('data', dataset)
train_data = np.memmap(os.path.join(data_dir, 'train.bin'), dtype=np.uint16, mode='r')
val_data = np.memmap(os.path.join(data_dir, 'val.bin'), dtype=np.uint16, mode='r')
def get_batch(split):
    data = train_data if split == 'train' else val_data
    ix = torch.randint(len(data) - block_size, (batch_size,))
    x = torch.stack([torch.from_numpy((data[i:i+block_size]).astype(np.int64)) for i in ix])
    y = torch.stack([torch.from_numpy((data[i+1:i+1+block_size]).astype(np.int64)) for i in ix])
    x, y = x.to(device), y.to(device)
    return x, y

# init these up here, can override if init_from='resume' (i.e. from a checkpoint)
iter_num = 0
best_val_loss = 1e9

# model init
model_args = dict(n_layer = n_layer, n_head = n_head, n_embd = n_embd, block_size = block_size, dropout = dropout)
if init_from == 'scratch':
    # init a new model from scratch
    print("Initializing a new model from scratch")
    gptconf = GPTConfig(**model_args)
    model = GPT(gptconf)
elif init_from == 'resume':
    print(f"Resuming training from {out_dir}")
    # resume training from a checkpoint.
    ckpt_path = os.path.join(out_dir, 'ckpt.pt')
    checkpoint = torch.load(ckpt_path, map_location=device)
    checkpoint_model_args = checkpoint['model_args']
    for k, v in model_args.items():
        assert checkpoint_model_args[k] == v, "for now"
        # TODO: think through how passed in params should interact with checkpoint params
    gptconf = GPTConfig(**model_args)
    model = GPT(gptconf)
    state_dict = checkpoint['model']
    # fix the keys of the state dictionary :(
    # honestly no idea how checkpoints sometimes get this prefix, have to debug more
    unwanted_prefix = '_orig_mod.'
    for k,v in list(state_dict.items()):
        if k.startswith(unwanted_prefix):
            state_dict[k[len(unwanted_prefix):]] = state_dict.pop(k)
    model.load_state_dict(state_dict)
    iter_num = checkpoint['iter_num']
    best_val_loss = checkpoint['best_val_loss']
elif init_from.startswith('gpt2'):
    print(f"Initializing from OpenAI GPT-2 weights: {init_from}")
    # initialize from OpenAI GPT-2 weights
    override_args = dict(dropout=dropout)
    model = GPT.from_pretrained(init_from, override_args)
    # read off and override the GPT sizing model args from the model config
    model_args['n_layer'] = model.config.n_layer
    model_args['n_head'] = model.config.n_head
    model_args['n_embd'] = model.config.n_embd
# crop down the model block size if desired
if block_size < model.config.block_size:
    model.crop_block_size(block_size)
model.to(device)

# optimizer
optimizer = model.configure_optimizers(weight_decay, learning_rate, betas)
if init_from == 'resume':
    optimizer.load_state_dict(checkpoint['optimizer'])

# compile the model
if compile:
    print("compiling the model... (takes a ~minute)")
    unoptimized_model = model
    model = torch.compile(model) # requires PyTorch 2.0

# wrap model into DDP container
if ddp:
    model = DDP(model, device_ids=[gpu_id])

@torch.no_grad()
def estimate_loss():
    out = {}
    model.eval()
    for split in ['train', 'val']:
        losses = torch.zeros(eval_iters)
        for k in range(eval_iters):
            X, Y = get_batch(split)
            with torch.amp.autocast(device_type="cuda", dtype=torch.bfloat16):
                logits, loss = model(X, Y)
            losses[k] = loss.item()
        out[split] = losses.mean()
    model.train()
    return out

# learning rate decay scheduler (cosine with warmup)
def get_lr(iter):
    # 1) linear warmup for warmup_iters steps
    if iter < warmup_iters:
        return learning_rate * iter / warmup_iters
    # 2) if iter > lr_decay_iters, return min learning rate
    if iter > lr_decay_iters:
        return min_lr
    # 3) in between, use cosine decay down to min learning rate
    decay_ratio = (iter - warmup_iters) / (lr_decay_iters - warmup_iters)
    assert 0 <= decay_ratio <= 1
    coeff = 0.5 * (1.0 + math.cos(math.pi * decay_ratio)) # coeff ranges 0..1
    return min_lr + coeff * (learning_rate - min_lr)

# logging
if wandb_log and gpu_id == 0:
    wandb.init(project=wandb_project, entity=wandb_entity, name=wandb_run_name)
    wandb.config = {
        "batch_size": batch_size,
        "block_size": block_size,
        "learning_rate": learning_rate, # TODO log everything else too
    }

# training loop
t0 = time.time()
while True:

    # determine the learning rate for this iteration
    if decay_lr:
        lr = get_lr(iter_num)
        for param_group in optimizer.param_groups:
            param_group['lr'] = lr
    else:
        lr = learning_rate

    if iter_num % eval_interval == 0 and gpu_id == 0:
        losses = estimate_loss()
        print(f"step {iter_num}: train loss {losses['train']:.4f}, val loss {losses['val']:.4f}")
        if wandb_log:
            wandb.log({
                "iter": iter_num,
                "train/loss": losses['train'],
                "val/loss": losses['val'],
                "lr": lr,
            })
        if losses['val'] < best_val_loss or always_save_checkpoint:
            best_val_loss = losses['val']
            raw_model = model.module if ddp else model
            if iter_num > 0:
                checkpoint = {
                    'model': raw_model.state_dict(),
                    'optimizer': optimizer.state_dict(),
                    'model_args': model_args,
                    'iter_num': iter_num,
                    'best_val_loss': best_val_loss,
                }
                print(f"saving checkpoint to {out_dir}")
                torch.save(checkpoint, os.path.join(out_dir, 'ckpt.pt'))
    if iter_num == 0 and eval_only:
        break

    X, Y = get_batch('train')
    with torch.amp.autocast(device_type="cuda", dtype=torch.bfloat16):
        logits, loss = model(X, Y)

    optimizer.zero_grad(set_to_none=True)
    loss.backward()
    # TODO: gradient clipping evaluate need for
    optimizer.step()

    t1 = time.time()
    dt = t1 - t0
    t0 = t1
    if iter_num % log_interval == 0 and gpu_id == 0:
        lossf = loss.item() # loss as float. TODO CPU-GPU sync: profile, make sure not slow af
        print(f"iter {iter_num}: loss {lossf:.4f}, time {dt*1000:.2f}ms")
    iter_num += 1

    # termination conditions
    if iter_num > max_iters:
        break

if ddp:
    destroy_process_group()