train_no527.py

"""
This file is used for experiments in excluding numbers or digits
"""

import os
import time
import math
import pickle
from contextlib import nullcontext

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
from main_utils import evaluate_text, evaluate_addition_new, get_encode_decode, evaluate_addition_multidigit

# -----------------------------------------------------------------------------
# default config values designed to train a gpt2 (124M) on OpenWebText
# I/O
out_dir = 'out'
resume_dir = None
eval_interval = 2000
log_interval = 1
eval_iters = 200
eval_only = False # if True, script exits right after the first eval
always_save_checkpoint = True # if True, always save a checkpoint after each eval
init_from = 'scratch' # 'scratch' or 'resume' or 'gpt2*'
# wandb logging
wandb_entity = 'ssdd'
wandb_log = True # disabled by default
wandb_project = 'owt'
wandb_run_name = 'gpt2' # 'run' + str(time.time())
# data
dataset = 'openwebtext'
gradient_accumulation_steps = 1 # used to simulate larger batch sizes
batch_size = 12 # if gradient_accumulation_steps > 1, this is the micro-batch size
block_size = 1024
train_data_path = 'train.bin'
val_data_path = 'val.bin'
multi_digit = False
num_digit = 5
binary = False
# using two data - data1 = text / data2 = addition
train_both = False # use seperate text/add data for train/val (get_batch uses this to sample from two differernt datasets)
data_ratio = 0.2 # ratio of data_path2 compared with data_path1
train_data_path2 = 'train_addition.bin' # only used when train_both = True
val_data_path2 = 'val_addition.bin'
# evaluation
eval_text = False # if True get perplexity using eval_text_data_path
eval_text_data_path = None # additional dataset to evaluate 
eval_addition = False # if True compute test accuracy of "a+b="
eval_addition_527 = True
start = None
start_527 = None
reverse_ab = False
reverse_c = False
zero_pad = False
algo_reason = False
# model
n_layer = 12
n_head = 12
n_embd = 768
dropout = 0.0 # for pretraining 0 is good, for finetuning try 0.1+
bias = False # do we use bias inside LayerNorm and Linear layers?
ckpt_path_name = 'ckpt.pt'
save_final = True
# adamw optimizer
learning_rate = 6e-4 # max learning rate
max_iters = 600000 # total number of training iterations
weight_decay = 1e-1
beta1 = 0.9
beta2 = 0.95
grad_clip = 1.0 # clip gradients at this value, or disable if == 0.0
# 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 = 600000 # should be ~= max_iters per Chinchilla
min_lr = None # minimum learning rate, should be ~= learning_rate/10 per Chinchilla
# DDP settings
backend = 'nccl' # 'nccl', 'gloo', etc.
# system
device = 'cuda' # examples: 'cpu', 'cuda', 'cuda:0', 'cuda:1' etc., or try 'mps' on macbooks
dtype = 'bfloat16' # 'float32', 'bfloat16', or 'float16', the latter will auto implement a GradScaler
compile = True # use PyTorch 2.0 to compile the model to be faster
use_flash = True
# -----------------------------------------------------------------------------
# import ipdb; ipdb.set_trace()
config_keys = [k for k,v in globals().items() if not k.startswith('_') and isinstance(v, (int, float, bool, str, type(None)))]
exec(open('configurator.py').read()) # overrides from command line or config file
config = {k: globals()[k] for k in config_keys} # will be useful for logging
# -----------------------------------------------------------------------------

if min_lr == None:
    min_lr = learning_rate/10

# various inits, derived attributes, I/O setup
ddp = int(os.environ.get('RANK', -1)) != -1 # is this a ddp run?
print('ddp: ', ddp)
if ddp:
    init_process_group(backend=backend)
    ddp_rank = int(os.environ['RANK'])
    ddp_local_rank = int(os.environ['LOCAL_RANK'])
    device = f'cuda:{ddp_local_rank}'
    torch.cuda.set_device(device)
    master_process = ddp_rank == 0 # this process will do logging, checkpointing etc.
    seed_offset = ddp_rank # each process gets a different seed
else:
    # if not ddp, we are running on a single gpu, and one process
    master_process = True
    seed_offset = 0

if master_process:
    os.makedirs(out_dir, exist_ok=True)
torch.manual_seed(1337 + seed_offset)
torch.backends.cuda.matmul.allow_tf32 = True # allow tf32 on matmul
torch.backends.cudnn.allow_tf32 = True # allow tf32 on cudnn
torch.backends.cudnn.benchmark = True # cudnn auto-tuner
torch.backends.cudnn.deterministic = False # cudnn auto-tuner

device_type = 'cuda' if 'cuda' in device else 'cpu' # for later use in torch.autocast
# note: float16 data type will automatically use a GradScaler
ptdtype = {'float32': torch.float32, 'bfloat16': torch.bfloat16, 'float16': torch.float16}[dtype]
ctx = nullcontext() if device_type == 'cpu' else torch.amp.autocast(device_type=device_type, dtype=ptdtype)

# poor man's data loader
data_dir = os.path.join('data', dataset)
train_data = np.memmap(os.path.join(data_dir, train_data_path), dtype=np.uint16, mode='r')
val_data = np.memmap(os.path.join(data_dir, val_data_path), dtype=np.uint16, mode='r')
if train_both:
    train_data2 = np.memmap(os.path.join(data_dir, train_data_path2), dtype=np.uint16, mode='r')
    val_data2 = np.memmap(os.path.join(data_dir, val_data_path2), dtype=np.uint16, mode='r')
if eval_text and eval_text_data_path: 
    eval_text_data = np.memmap(eval_text_data_path, dtype=np.uint16, mode='r')

def get_batch(split):
    data = train_data if split == 'train' else val_data
    if train_both:
        data2 = train_data2 if split == 'train' else val_data2
        batch_size2 = int(batch_size*data_ratio)
        ix = torch.randint(len(data) - block_size, (batch_size-batch_size2,))
        ix2 = torch.randint(len(data2) - block_size, (batch_size2,))
    else:
        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])
    if train_both:
        x2 = torch.stack([torch.from_numpy((data2[i:i+block_size]).astype(np.int64)) for i in ix2])
        y2 = torch.stack([torch.from_numpy((data2[i+1:i+1+block_size]).astype(np.int64)) for i in ix2])
        x = torch.cat([x,x2])
        y = torch.cat([y,y2])    

    if device_type == 'cuda':
        # pin arrays x,y, which allows us to move them to GPU asynchronously (non_blocking=True)
        x, y = x.pin_memory().to(device, non_blocking=True), y.pin_memory().to(device, non_blocking=True)
    else:
        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
best_perplexity = 1e9 # on text data
best_accuracy = -1 # on addition data

# attempt to derive vocab_size from the dataset
meta_path = os.path.join(data_dir, 'meta.pkl')
meta_vocab_size = None
if os.path.exists(meta_path):
    with open(meta_path, 'rb') as f:
        meta = pickle.load(f)
    meta_vocab_size = meta['vocab_size']
    print(f"found vocab_size = {meta_vocab_size} (inside {meta_path})")
else:
    meta_path = None

# model init
model_args = dict(n_layer=n_layer, n_head=n_head, n_embd=n_embd, block_size=block_size,
                  bias=bias, vocab_size=None, dropout=dropout, use_flash=use_flash) # start with model_args from command line
if init_from == 'scratch':
    # init a new model from scratch
    print("Initializing a new model from scratch")
    # determine the vocab size we'll use for from-scratch training
    if meta_vocab_size is None:
        print("defaulting to vocab_size of GPT-2 to 50304 (50257 rounded up for efficiency)")
    model_args['vocab_size'] = meta_vocab_size if meta_vocab_size is not None else 50304
    gptconf = GPTConfig(**model_args)
    model = GPT(gptconf)
elif init_from == 'resume':
    if resume_dir:
        checkpoint = torch.load(resume_dir, map_location=device)
    else:
        print(f"Resuming training from {out_dir}")
        # resume training from a checkpoint.
        ckpt_path = os.path.join(out_dir, ckpt_path_name)
        checkpoint = torch.load(ckpt_path, map_location=device)
    checkpoint_model_args = checkpoint['model_args']
    # force these config attributes to be equal otherwise we can't even resume training
    # the rest of the attributes (e.g. dropout) can stay as desired from command line
    for k in ['n_layer', 'n_head', 'n_embd', 'block_size', 'bias', 'vocab_size']:
        model_args[k] = checkpoint_model_args[k]
    # create the model
    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']
    max_iters += iter_num
    best_val_loss = checkpoint['best_val_loss']
    if 'best_perplexity' in checkpoint.keys(): 
        best_perplexity = checkpoint['best_perplexity']
    if 'best_accuracy' in checkpoint.keys():
        best_accuracy = checkpoint['best_accuracy']
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 the created config params, so we can store them into checkpoint correctly
    for k in ['n_layer', 'n_head', 'n_embd', 'block_size', 'bias', 'vocab_size']:
        model_args[k] = getattr(model.config, k)
# crop down the model block size if desired, using model surgery
if block_size < model.config.block_size:
    model.crop_block_size(block_size)
    model_args['block_size'] = block_size # so that the checkpoint will have the right value
model.to(device)

# initialize a GradScaler. If enabled=False scaler is a no-op
scaler = torch.cuda.amp.GradScaler(enabled=(dtype == 'float16'))

# optimizer
optimizer = model.configure_optimizers(weight_decay, learning_rate, (beta1, beta2), device_type)
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=[ddp_local_rank])

# helps estimate an arbitrarily accurate loss over either split using many batches
@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 ctx:
                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(it):
    # 1) linear warmup for warmup_iters steps
    if it < warmup_iters:
        return learning_rate * it / warmup_iters
    # 2) if it > lr_decay_iters, return min learning rate
    if it > lr_decay_iters:
        return min_lr
    # 3) in between, use cosine decay down to min learning rate
    decay_ratio = (it - 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 master_process:
    import wandb
    wandb.init(project=wandb_project, entity=wandb_entity, name=wandb_run_name, config=config)

# training loop
X, Y = get_batch('train') # fetch the very first batch
t0 = time.time()
local_iter_num = 0 # number of iterations in the lifetime of this process
raw_model = model.module if ddp else model # unwrap DDP container if needed
running_mfu = -1.0
encode, decode = get_encode_decode(meta_path)

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

    # evaluate the loss on train/val sets and write checkpoints
    if iter_num % eval_interval == 0 and master_process:
        losses = estimate_loss()
        if eval_text and eval_text_data_path:
            ppl = evaluate_text(config, model, eval_text_data, ctx)
            print(f"perplexity of evluation text data: {ppl}")
        if eval_addition:
            if multi_digit:
                config['start'] = start
                digit_accuracy_dictionary = evaluate_addition_multidigit(config, model, ctx, encode, decode, verbose=False, num_digit=num_digit, zero_pad=zero_pad, reverse_ab=reverse_ab, reverse_c=reverse_c, algo_reason=algo_reason, binary=binary)
            test_accuracy, _ = evaluate_addition_new(config, model, ctx, encode, decode, verbose=True, num_digit=num_digit, zero_pad=zero_pad, reverse_ab=reverse_ab, reverse_c=reverse_c, algo_reason=algo_reason, binary=binary)
        if eval_addition_527:
            config['start'] = start_527
            test_accuracy_527, _ = evaluate_addition_new(config, model, ctx, encode, decode, verbose=True, num_digit=num_digit, zero_pad=zero_pad, reverse_ab=reverse_ab, reverse_c=reverse_c, algo_reason=algo_reason, binary=binary)
        print(f"step {iter_num}: train loss {losses['train']:.4f}, val loss {losses['val']:.4f}")
        print(f"step {iter_num}: test acc {test_accuracy:.4f}, 527 test acc {test_accuracy_527:.4f}")
        if wandb_log:
            wandb_dict = {
                "iter": iter_num,
                "train/loss": losses['train'],
                "val/loss": losses['val'],
                "lr": lr,
                "mfu": running_mfu*100, # convert to percentage,
                "ppl": ppl if (eval_text and eval_text_data_path) else None, 
                "test/accuracy": test_accuracy if eval_addition else None,
                "test/accuracy_527": test_accuracy_527 if eval_addition_527 else None
            }
            if multi_digit: wandb_dict.update(digit_accuracy_dictionary)
            wandb.log(wandb_dict)
        checkpoint = {
                    'model': raw_model.state_dict(),
                    'optimizer': optimizer.state_dict(),
                    'model_args': model_args,
                    'iter_num': iter_num,
                    'best_val_loss': best_val_loss,
                    'best_perplexity': best_perplexity,
                    'best_accuracy': best_accuracy,
                    'config': config,
                }
        if losses['val'] < best_val_loss or always_save_checkpoint:
            best_val_loss = losses['val']
            checkpoint['best_val_loss'] = best_val_loss
            if iter_num > 0:
                print(f"saving checkpoint to {out_dir}/{ckpt_path_name}")
                torch.save(checkpoint, os.path.join(out_dir, ckpt_path_name))
        if eval_addition and test_accuracy > best_accuracy:
            best_accuracy = test_accuracy
            checkpoint['best_accuracy'] = best_accuracy
            if iter_num > 0:
                print(f"saving checkpoint to {out_dir}/{ckpt_path_name}")
                torch.save(checkpoint, os.path.join(out_dir, ckpt_path_name.split('.pt')[0]+'_acc.pt'))
        
    if iter_num == 0 and eval_only:
        break

    # forward backward update, with optional gradient accumulation to simulate larger batch size
    # and using the GradScaler if data type is float16
    for micro_step in range(gradient_accumulation_steps):
        if ddp:
            # in DDP training we only need to sync gradients at the last micro step.
            # the official way to do this is with model.no_sync() context manager, but
            # I really dislike that this bloats the code and forces us to repeat code
            # looking at the source of that context manager, it just toggles this variable
            model.require_backward_grad_sync = (micro_step == gradient_accumulation_steps - 1)
        with ctx:
            logits, loss = model(X, Y)
        # immediately async prefetch next batch while model is doing the forward pass on the GPU
        X, Y = get_batch('train')
        # backward pass, with gradient scaling if training in fp16
        scaler.scale(loss).backward()
    # clip the gradient
    if grad_clip != 0.0:
        scaler.unscale_(optimizer)
        torch.nn.utils.clip_grad_norm_(model.parameters(), grad_clip)
    # step the optimizer and scaler if training in fp16
    scaler.step(optimizer)
    scaler.update()
    # flush the gradients as soon as we can, no need for this memory anymore
    optimizer.zero_grad(set_to_none=True)

    # timing and logging
    t1 = time.time()
    dt = t1 - t0
    t0 = t1
    if iter_num % log_interval == 0 and master_process:
        lossf = loss.item() # loss as float. note: this is a CPU-GPU sync point
        if local_iter_num >= 5: # let the training loop settle a bit
            mfu = raw_model.estimate_mfu(batch_size * gradient_accumulation_steps, dt)
            running_mfu = mfu if running_mfu == -1.0 else 0.9*running_mfu + 0.1*mfu
        print(f"iter {iter_num}: loss {lossf:.4f}, time {dt*1000:.2f}ms, mfu {running_mfu*100:.2f}%")
    iter_num += 1
    local_iter_num += 1

    # termination conditions
    if iter_num > max_iters:
        break

if save_final:
    print(f"saving final checkpoint to {out_dir}")
    torch.save(checkpoint, os.path.join(out_dir, ckpt_path_name.split('.pt')[0]+'_final.pt'))

if ddp:
    destroy_process_group()