train_sophiag.py

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

import numpy as np
import torch
import torch.nn.functional as F
from torch.nn.parallel import DistributedDataParallel as DDP
from torch.distributed import init_process_group, destroy_process_group
from model import GPTConfig, GPT
import torch.autograd as autograd

# -----------------------------------------------------------------------------
# default config values designed to train a gpt2 (124M) on OpenWebText
# I/O
out_dir = 'out'
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_log = False # disabled by default
wandb_project = 'owt'
wandb_run_name = 'gpt2' # 'run' + str(time.time())
# data
dataset = 'openwebtext'
gradient_accumulation_steps = 5 # used to simulate larger batch sizes
batch_size = 12 # if gradient_accumulation_steps > 1, this is the micro-batch size
block_size = 1024
total_bs = 480
# 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?
# optimizer
optimizer_name = 'sophiag' 
learning_rate = 3e-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
rho = 0.03
interval = 10
hess_interval = interval
variant = 4 
# 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 = 1.5e-5 # 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'
compile = True # use PyTorch 2.0 to compile the model to be faster
scale_attn_by_inverse_layer_idx = True
# -----------------------------------------------------------------------------
config_keys = [k for k,v in globals().items() if not k.startswith('_') and isinstance(v, (int, float, bool, str))]
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
# -----------------------------------------------------------------------------

# various inits, derived attributes, I/O setup
ddp = int(os.environ.get('RANK', -1)) != -1 # is this a ddp run?
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
    ddp_rank = 0                             #ddp_rank is used in get_batch function so this has to be here also when running locally
    master_process = True
    seed_offset = 0
    gradient_accumulation_steps *= 8 # simulate 8 gpus

if master_process:
    os.makedirs(out_dir, exist_ok=True)
torch.manual_seed(2099)
torch.backends.cuda.matmul.allow_tf32 = True # allow tf32 on matmul
torch.backends.cudnn.allow_tf32 = True # allow tf32 on cudnn
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.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_list = []
    for jj in range(10):
        ix_list.append(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_list[ddp_rank]])
    y = torch.stack([torch.from_numpy((data[i+1:i+1+block_size]).astype(np.int64)) for i in ix_list[ddp_rank]])
    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

# 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})")

# 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, scale_attn_by_inverse_layer_idx=scale_attn_by_inverse_layer_idx) # 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':
    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']
    # 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']
    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 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)


optimizer = model.configure_optimizers(optimizer_name, weight_decay, learning_rate, (beta1, beta2), rho, device_type)
if init_from == 'resume':
    optimizer.load_state_dict(checkpoint['optimizer'])
    del state_dict
    del checkpoint
# 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, 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
num_param = 1
num_effective = 0
momentum_norm = 0
hessian_norm = 0
hessian_norm2 = 0
clip_time = 0
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()
        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,
                "mfu": running_mfu*100, # convert to percentage
            }, step=iter_num)
        if losses['val'] < best_val_loss or always_save_checkpoint:
            best_val_loss = losses['val']
            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,
                    'config': config,
                }
                print(f"saving checkpoint to {out_dir}")
                torch.save(checkpoint, os.path.join(out_dir, 'ckpt.pt'))
        if iter_num % (eval_interval * 5) == 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,
                'config': config,
            }
            print(f"saving checkpoint to {out_dir}")
            torch.save(checkpoint, os.path.join(out_dir, 'ckpt_'+str(iter_num)+'.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
    if iter_num % hess_interval != hess_interval - 1:
        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
            (loss / gradient_accumulation_steps).backward()
        # clip the gradient
        if grad_clip != 0.0:
            total_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), grad_clip)
            if total_norm.item() > grad_clip:
                clip_time += 1
        # step the optimizer and scaler if training in fp16
        optimizer.step(bs=total_bs * block_size)
        # 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}%")
            total_param_norm = 0
            momentum_norm = 0
            params = []
            for (name, p) in model.named_parameters():
                params.append(p)
            for p in params:
                param_norm = p.data.norm(2)
                total_param_norm += param_norm.item() ** 2
            total_param_norm = total_param_norm ** 0.5
            momentum_norm = 0
            LL = len(optimizer.state_dict()['state'])
            for jj in range(LL):
                momentum_norm += (optimizer.state_dict()['state'][jj]['exp_avg'].detach().norm(2)) ** 2
            momentum_norm = torch.sqrt(momentum_norm).item()
            if wandb_log:
                wandb.log({
                    "iter": iter_num,
                    "train/loss": lossf,
                    "lr": lr,
                    "param_norm": total_param_norm,
                    "momentum_norm" : momentum_norm,
                    "hessian_norm": hessian_norm,
                    "hessian_norm2": hessian_norm2,
                    "train/win_rate": num_effective / num_param,
                    "train/clip_rate": clip_time / (iter_num + 1)
                    
                }, step=iter_num)
        iter_num += 1
        local_iter_num += 1
    else:
        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
            (loss / gradient_accumulation_steps).backward()
        # clip the gradient
        if grad_clip != 0.0:
            total_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), grad_clip)
            if total_norm.item() > grad_clip:
                clip_time += 1
        # step the optimizer and scaler if training in fp16
        optimizer.step(bs=total_bs * block_size)
        # 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}%")

            total_param_norm = 0
            momentum_norm = 0
                
            if wandb_log:
                wandb.log({
                    "iter": iter_num,
                    "train/loss": lossf,
                    "lr": lr,
                    "param_norm": total_param_norm,
                    "momentum_norm" : momentum_norm,
                    "hessian_norm": hessian_norm.item(),
                    "hessian_norm2": hessian_norm2,
                    "train/win_rate": num_effective / num_param,
                    "train/clip_rate": clip_time / (iter_num + 1)
                    
                }, step=iter_num)
        iter_num += 1
        local_iter_num += 1

        for micro_step in range(gradient_accumulation_steps):
            if ddp:
                model.require_backward_grad_sync = (micro_step == gradient_accumulation_steps - 1)
            with ctx:
                logits, _ = model(X, 0)
            X, Y = get_batch('train')
            samp_dist = torch.distributions.Categorical(logits=logits)
            y_sample = samp_dist.sample()
            loss = F.cross_entropy(logits.view(-1, logits.size(-1)), y_sample.view(-1), ignore_index=-1)
            # backward pass, with gradient scaling if training in fp16
            (loss / gradient_accumulation_steps).backward()
        # clip the gradient
        if grad_clip != 0.0:
            total_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), grad_clip)
        # step the optimizer and scaler if training in fp16
        optimizer.update_hessian()
        # flush the gradients as soon as we can, no need for this memory anymore
        optimizer.zero_grad(set_to_none=True)

        num_param = 0
        num_effective = 0
        hessian_norm = 0
        hessian_norm2 = 0
        
        LL = len(optimizer.state_dict()['state'])
        
        for jj in range(LL):
            num_param += optimizer.state_dict()['state'][jj]['exp_avg'].numel()
            num_effective += torch.sum(torch.abs(optimizer.state_dict()['state'][jj]['exp_avg']) < rho * total_bs * block_size * optimizer.state_dict()['state'][jj]['hessian'])
            hessian_norm += optimizer.state_dict()['state'][jj]['hessian'].detach().norm(1).item()
            hessian_norm2 += optimizer.state_dict()['state'][jj]['hessian'].detach().norm(2).item() ** 2
        hessian_norm2 = hessian_norm2 ** 0.5

        
        
        t1 = time.time()
        dt = t1 - t0
        t0 = t1
        if master_process:
            # 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}: time {dt*1000:.2f}ms, mfu {running_mfu*100:.2f}%")


    # termination conditions
    if iter_num > max_iters:
        break

if ddp:
    destroy_process_group()