train_simple.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 on a single GPU, example:

$ python train.py --batch_size=32 --compile=False



To run with DDP on 4 gpus on 1 node, example:

$ torchrun --standalone --nproc_per_node=4 train.py



To run with DDP on 4 gpus across 2 nodes, example:

- Run on the first (master) node with example IP 123.456.123.456:

$ torchrun --nproc_per_node=8 --nnodes=2 --node_rank=0 --master_addr=123.456.123.456 --master_port=1234 train.py

- Run on the worker node:

$ torchrun --nproc_per_node=8 --nnodes=2 --node_rank=1 --master_addr=123.456.123.456 --master_port=1234 train.py

(If your cluster does not have Infiniband interconnect prepend NCCL_IB_DISABLE=1)

"""

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


import numpy as np
import torch
from torch.nn.parallel import DistributedDataParallel as DDP
from torch.distributed import init_process_group, destroy_process_group
import networkx as nx
import re 

from model.transformer import GPTConfig, GPT
from logger import get_logger
import logging

# Suppress verbose PyTorch Dynamo debug messages
logging.getLogger("torch._dynamo").setLevel(logging.WARNING)

# -----------------------------------------------------------------------------
# the input parameters

parser = argparse.ArgumentParser(description='Training of the NanoGPT.')

parser.add_argument('--dataset', type=str, default='simple_graph', help='Name of the dataset to use')  
parser.add_argument('--n_layer', type=int, default=1, help='Number of layers (default: 1)')  
parser.add_argument('--n_head', type=int, default=1, help='Number of attention heads (default: 1)')  
parser.add_argument('--n_embd', type=int, default=120, help='Size of the embeddings (default: 120)')
parser.add_argument('--max_iters', type=int, default=10000, help='Number of Iterations (default: 10000)')
parser.add_argument('--num_nodes', type=int, default=100, help='Number of Nodes (default: 100)')
parser.add_argument('--num_of_paths', type=int, default=20, help='Number of Paths (default: 1)')
parser.add_argument('--init_ckpt', type=int, default=0, help='Initial checkpoint iteration to resume from (default: 0 means train from scratch)')


args = parser.parse_args()

dataset = args.dataset
n_layer = args.n_layer
n_head = args.n_head
n_embd = args.n_embd
max_iters = args.max_iters
num_nodes = args.num_nodes
num_of_paths = args.num_of_paths
init_ckpt = args.init_ckpt

data_dir = os.path.join('data', f'{dataset}/{num_nodes}')
with open(os.path.join(data_dir, 'meta.pkl'), 'rb') as f:
    meta = pickle.load(f)
    
stoi, itos = meta['stoi'], meta['itos']
block_size = meta['block_size']

out_dir = f'out/{dataset}_{n_layer}_{n_head}_{n_embd}_{num_nodes}'

# -----------------------------------------------------------------------------
# default config values designed to train a gpt2 (124M) on OpenWebText
# I/O
eval_interval = max_iters // 10
log_interval = max_iters // 100
eval_iters = max_iters // 10

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 = 'resume' if init_ckpt > 0 else 'scratch' # determined by --init_ckpt argument
# wandb logging
wandb_log = False # disabled by default
wandb_project = 'owt'
wandb_run_name = 'gpt2' # 'run' + str(time.time())
# data
#dataset = 'reasoning'
gradient_accumulation_steps = 1 # used to simulate larger batch sizes
train_batch_size = 1024 # if gradient_accumulation_steps > 1, this is the micro-batch size
val_batch_size = 64
batch_size = train_batch_size
#block_size = 64
# model
#n_layer = 1 #12
#n_head = 1 #12
#n_embd = 384 #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?
# adamw optimizer
learning_rate = 5e-4 # max learning rate 
#max_iters = 50000 # 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 = max_iters//20 # how many steps to warm up for
lr_decay_iters = max_iters # should be ~= max_iters per Chinchilla
min_lr = learning_rate/10 # 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

'''check_type = 'shortest'

max_path_len = 10

max_new_tokens = 200

flag = 0

test_interval = 100'''
# -----------------------------------------------------------------------------
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'])
    ddp_world_size = int(os.environ['WORLD_SIZE'])
    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
    assert gradient_accumulation_steps % torch.cuda.device_count() == 0
    gradient_accumulation_steps //= torch.cuda.device_count()
else:
    # if not ddp, we are running on a single gpu, and one process
    master_process = True
    seed_offset = 0
    ddp_world_size = 1
tokens_per_iter = gradient_accumulation_steps * ddp_world_size * batch_size * block_size
print(f"tokens per iteration will be: {tokens_per_iter:,}")

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
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
if(num_of_paths == 0):
    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')
else:
    train_data = np.memmap(os.path.join(data_dir, f'train_{num_of_paths}.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
    batch_size = train_batch_size if split == 'train' else val_batch_size
    
    data_size = block_size + 1
    data = train_data if split == 'train' else val_data
    ix = torch.randint( (len(data) - data_size)//data_size , (batch_size,)) * data_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 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

# logger
if(num_of_paths == 0):
    logger = get_logger(os.path.join(out_dir, "no_output_train.log"))
    log_file_name = os.path.join(out_dir, "train.log")
    #logger.setLevel(logging.DEBUG)
else:
    logger = get_logger(os.path.join(out_dir, f'no_output_train_{num_of_paths}.log'))
    log_file_name = os.path.join(out_dir, f"train_{num_of_paths}.log")
    #logger.setLevel(logging.DEBUG)



# 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})")

def get_shortest(p_graph):
    shortest_paths = {}
    for i in p_graph.nodes:
        for j in p_graph.nodes:
            try:
                shortest_paths[(i,j)] = list(nx.all_shortest_paths(p_graph, i, j))
            except:
                shortest_paths[(i,j)] = []
    return shortest_paths

if dataset == 'reasoning':
    p_graph_path = os.path.join(data_dir, 'fixed_model.graphml')
    p_graph = nx.read_graphml(p_graph_path)
    shortest_paths = get_shortest(p_graph)
    
stoi, itos = meta['stoi'], meta['itos']
decode = lambda l: ''.join([itos[i] for i in l])

# 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) # start with model_args from command line
if init_from == 'scratch':
    print("Initializing a new model from scratch")
    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':
    # Determine the checkpoint file path based on init_ckpt and num_of_paths
    if num_of_paths == 0:
        ckpt_path = os.path.join(out_dir, f'{init_ckpt}_ckpt.pt')
    else:
        ckpt_path = os.path.join(out_dir, f'{init_ckpt}_ckpt_{num_of_paths}.pt')
    print(f"Resuming training from {ckpt_path}")
    # resume training from a checkpoint.
    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}")
    override_args = dict(dropout=dropout)
    model = GPT.from_pretrained(init_from, override_args)
    for k in ['n_layer', 'n_head', 'n_embd', 'block_size', 'bias', 'vocab_size']:
        model_args[k] = getattr(model.config, k)

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
optimizer = model.configure_optimizers(weight_decay, learning_rate, (beta1, beta2), device_type)
if init_from == 'resume':
    optimizer.load_state_dict(checkpoint['optimizer'])

# initialize a GradScaler. If enabled=False scaler is a no-op
scaler = torch.cuda.amp.GradScaler(enabled=(dtype == 'float16'))
if init_from == 'resume' and 'scaler' in checkpoint:
    scaler.load_state_dict(checkpoint['scaler'])

# restore random states for reproducibility
if init_from == 'resume':
    if 'torch_rng_state' in checkpoint:
        # RNG state must be a ByteTensor on CPU
        torch.set_rng_state(checkpoint['torch_rng_state'].cpu())
    if 'cuda_rng_state' in checkpoint and checkpoint['cuda_rng_state'] is not None and torch.cuda.is_available():
        # CUDA RNG state must also be a ByteTensor on CPU before setting
        torch.cuda.set_rng_state(checkpoint['cuda_rng_state'].cpu())
    if 'numpy_rng_state' in checkpoint:
        np.random.set_state(checkpoint['numpy_rng_state'])

checkpoint = None # free up memory

# 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:
                _, 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)

def open_and_append(filename, text):
    with open(filename, 'a') as file:
        file.write(text + '\n')

# 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
accuracy = []
corrects = []
totals = []
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}")
        logger.info(f"step {iter_num}: train loss {losses['train']:.4f}, val loss {losses['val']:.4f}")
        open_and_append(log_file_name, 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
            })
        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,
                    'scaler': scaler.state_dict(),
                    'torch_rng_state': torch.get_rng_state(),
                    'cuda_rng_state': torch.cuda.get_rng_state() if torch.cuda.is_available() else None,
                    'numpy_rng_state': np.random.get_state(),
                }
                print(f"saving checkpoint to {out_dir}")
                logger.info(f"saving checkpoint to {out_dir}")
                open_and_append(log_file_name, "saving checkpoint to {out_dir}")
                if(num_of_paths == 0):
                    torch.save(checkpoint, os.path.join(out_dir, f'{iter_num}_ckpt.pt'))
                else:
                    torch.save(checkpoint, os.path.join(out_dir, f'{iter_num}_ckpt_{num_of_paths}.pt'))

    # if iter_num % test_interval == 0 and master_process:
    #     correct, tot = test_model()
    #     corrects.append(correct)
    #     totals.append(tot)

    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:
            model.require_backward_grad_sync = (micro_step == gradient_accumulation_steps - 1)
        with ctx:
            logits, loss = model(X, Y)
            loss = loss / gradient_accumulation_steps # scale the loss to account for gradient accumulation
        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)
    scaler.step(optimizer)
    scaler.update()
    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() * gradient_accumulation_steps
        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}%")
        logger.info(f"iter {iter_num}: loss {lossf:.4f}")
        open_and_append(log_file_name, f"iter {iter_num}: loss {lossf:.4f}")
    iter_num += 1
    local_iter_num += 1

    if iter_num > max_iters:
        break

torch.save(torch.tensor(corrects).cpu(), os.path.join(out_dir, f'corrects.pt'))
torch.save(torch.tensor(totals).cpu(), os.path.join(out_dir, f'totals.pt'))

if ddp:
    destroy_process_group()