llama-recipes/src/llama_recipes/utils/train_utils.py

# Copyright (c) Meta Platforms, Inc. and affiliates.
# This software may be used and distributed according to the terms of the Llama 2 Community License Agreement.

import os
import time
import yaml
from pathlib import Path
from pkg_resources import packaging
import contextlib
import gc

import torch
import torch.cuda.nccl as nccl
import torch.distributed as dist
from torch.distributed.fsdp import StateDictType
from torch.distributed.fsdp.sharded_grad_scaler import ShardedGradScaler
# from torch.utils.flop_counter import FlopCounterMode
from tqdm import tqdm
from transformers import LlamaTokenizer


from llama_recipes.model_checkpointing import save_model_checkpoint, save_model_and_optimizer_sharded, save_optimizer_checkpoint
from llama_recipes.policies import fpSixteen,bfSixteen_mixed, get_llama_wrapper
from llama_recipes.utils.memory_utils import MemoryTrace
from llama_recipes.utils.tflop_counter import FlopCounterMode

@contextlib.contextmanager
def maybe_run_profiler(cfg, *args, **kwargs):
    use_profiler: bool = cfg.profiler

    if use_profiler:
        with torch.profiler.profile(
            activities=[
                torch.profiler.ProfilerActivity.CPU,
                torch.profiler.ProfilerActivity.CUDA,
            ],
            schedule=torch.profiler.schedule(wait=1, warmup=2, active=3, repeat=1),
            on_trace_ready=torch.profiler.tensorboard_trace_handler(
                cfg.profile_output_dir
            ),
            profile_memory=True,
            with_stack=False,
            record_shapes=True,
        ) as torch_profiler:
            yield torch_profiler
    else:
        torch_profiler = contextlib.nullcontext()
        yield None
            
def get_total_flops(model):
    return (sum([v for _, v in model.flop_counts["Global"].items()]))

def set_tokenizer_params(tokenizer: LlamaTokenizer):
    tokenizer.pad_token_id = 0
    tokenizer.padding_side = "left"
    
# Converting Bytes to Megabytes
def byte2mb(x):
    return int(x / 2**20)

def train(model, train_dataloader,eval_dataloader, tokenizer, optimizer, lr_scheduler, gradient_accumulation_steps, train_config, fsdp_config=None, local_rank=None, rank=None):
    """
    Trains the model on the given dataloader
    
    Args:
        model: The model to be trained
        train_dataloader: The dataloader containing the training data
        optimizer: The optimizer used for training
        lr_scheduler: The learning rate scheduler
        gradient_accumulation_steps: The number of steps to accumulate gradients before performing a backward/update operation
        num_epochs: The number of epochs to train for
        local_rank: The rank of the current node in a distributed setting
        train_config: The training configuration
        eval_dataloader: The dataloader containing the eval data
        tokenizer: tokenizer used in the eval for decoding the predicitons
    
    Returns: results dictionary containing average training and validation perplexity and loss
    """
    # Create a gradient scaler for fp16
    if train_config.use_fp16 and train_config.enable_fsdp:
        scaler = ShardedGradScaler()
    elif train_config.use_fp16 and not train_config.enable_fsdp:
        scaler = torch.cuda.amp.GradScaler() 
    if train_config.enable_fsdp:
        world_size = int(os.environ["WORLD_SIZE"]) 
    train_prep = []
    train_loss = []
    val_prep = []
    val_loss =[]
    epoch_times = []
    checkpoint_times = []
    results = {}
    best_val_loss = float("inf")
    for epoch in range(train_config.num_epochs):
        epoch_start_time = time.perf_counter()
        with MemoryTrace() as memtrace:  # track the memory usage
            model.train()
            total_loss = 0.0
            total_length = len(train_dataloader)//gradient_accumulation_steps
            pbar = tqdm(colour="blue", desc=f"Training Epoch: {epoch+1}", total=total_length, dynamic_ncols=True)
            with maybe_run_profiler(train_config) as torch_profiler:
                for step, batch in enumerate(train_dataloader):
                    if step > 5:
                        break
                    gc.collect(1)
                    for key in batch.keys():
                        if train_config.enable_fsdp:
                            batch[key] = batch[key].to(local_rank)
                        else:
                            batch[key] = batch[key].to('cuda:0') 
                    flop_check_done = False 
                    if train_config.flop_counter and  step == 3 and not flop_check_done:
                        flop_counter = FlopCounterMode(rank=local_rank)
                        with flop_counter:           
                            loss = model(**batch).loss
                            loss = loss / gradient_accumulation_steps
                            total_loss += loss.detach().float()
                            if train_config.use_fp16:
                                # if fp16 is enabled, use gradient scaler to handle gradient update
                                scaler.scale(loss).backward()
                                if (step + 1) % gradient_accumulation_steps == 0 or step == len(train_dataloader) - 1:
                                    scaler.step(optimizer)
                                    scaler.update()
                                    optimizer.zero_grad()
                                    pbar.update(1)
                            else:
                                # regular backpropagation when fp16 is not used
                                loss.backward()
                                TFlops = get_total_flops(flop_counter) / 1e12
                                flop_check_done = True
                                if (step + 1) % gradient_accumulation_steps == 0 or step == len(train_dataloader) - 1:
                                    optimizer.step()
                                    optimizer.zero_grad()
                                    pbar.update(1)
                        
                    else:
                        loss = model(**batch).loss
                        loss = loss / gradient_accumulation_steps
                        total_loss += loss.detach().float()
                        if train_config.use_fp16:
                            # if fp16 is enabled, use gradient scaler to handle gradient update
                            scaler.scale(loss).backward()
                            if (step + 1) % gradient_accumulation_steps == 0 or step == len(train_dataloader) - 1:
                                scaler.step(optimizer)
                                scaler.update()
                                optimizer.zero_grad()
                                pbar.update(1)
                        else:
                            # regular backpropagation when fp16 is not used
                            loss.backward()
                            if (step + 1) % gradient_accumulation_steps == 0 or step == len(train_dataloader) - 1:
                                optimizer.step()
                                optimizer.zero_grad()
                                pbar.update(1)
                    pbar.set_description(f"Training Epoch: {epoch+1}/{train_config.num_epochs}, step {step}/{len(train_dataloader)} completed (loss: {loss.detach().float()})")
                pbar.close()
                
        epoch_end_time = time.perf_counter()-epoch_start_time
        epoch_times.append(epoch_end_time)    
        # Reducing total_loss across all devices if there's more than one CUDA device
        if torch.cuda.device_count() > 1 and train_config.enable_fsdp:
            dist.all_reduce(total_loss, op=dist.ReduceOp.SUM)
        train_epoch_loss = total_loss / len(train_dataloader)
        if train_config.enable_fsdp:
            train_epoch_loss = train_epoch_loss/world_size
        train_perplexity = torch.exp(train_epoch_loss)
        
        train_prep.append(train_perplexity)
        train_loss.append(train_epoch_loss)
        
        if train_config.enable_fsdp:
            if rank==0:
                print(f"Max CUDA memory allocated was {memtrace.peak} GB")
                print(f"Max CUDA memory reserved was {memtrace.max_reserved} GB")
                print(f"Peak active CUDA memory was {memtrace.peak_active_gb} GB")
                print(f"Cuda Malloc retires : {memtrace.cuda_malloc_retires}")
                print(f"CPU Total Peak Memory consumed during the train (max): {memtrace.cpu_peaked + memtrace.cpu_begin} GB")
        else:
            print(f"Max CUDA memory allocated was {memtrace.peak} GB")
            print(f"Max CUDA memory reserved was {memtrace.max_reserved} GB")
            print(f"Peak active CUDA memory was {memtrace.peak_active_gb} GB")
            print(f"Cuda Malloc retires : {memtrace.cuda_malloc_retires}")
            print(f"CPU Total Peak Memory consumed during the train (max): {memtrace.cpu_peaked + memtrace.cpu_begin} GB")
        
        # Update the learning rate as needed
        lr_scheduler.step()
          
        if train_config.run_validation:
            eval_ppl, eval_epoch_loss = evaluation(model, train_config, eval_dataloader, local_rank, tokenizer)
            checkpoint_start_time = time.perf_counter()
            if train_config.save_model and eval_epoch_loss < best_val_loss:
                if train_config.enable_fsdp:
                    dist.barrier()
                if train_config.use_peft:
                    if train_config.enable_fsdp:
                        if rank==0:
                            print(f"we are about to save the PEFT modules")
                    else:
                        print(f"we are about to save the PEFT modules")
                    model.save_pretrained(train_config.output_dir)  
                    if train_config.enable_fsdp:
                        if rank==0: 
                            print(f"PEFT modules are saved in {train_config.output_dir} directory")
                    else:
                        print(f"PEFT modules are saved in {train_config.output_dir} directory")
                        
                else:
                    if not train_config.use_peft and fsdp_config.checkpoint_type == StateDictType.FULL_STATE_DICT:
                        
                        save_model_checkpoint(
                            model, optimizer, rank, train_config, epoch=epoch
                        )
                    elif not train_config.use_peft and fsdp_config.checkpoint_type == StateDictType.SHARDED_STATE_DICT:
                        print(" Saving the FSDP model checkpoints using SHARDED_STATE_DICT")
                        print("=====================================================")
                        
                        save_model_and_optimizer_sharded(model, rank, train_config)
                        if train_config.save_optimizer:
                            save_model_and_optimizer_sharded(model, rank, train_config, optim=optimizer)
                            print(" Saving the FSDP model checkpoints and optimizer using SHARDED_STATE_DICT")
                            print("=====================================================")

                    if not train_config.use_peft and  train_config.save_optimizer:
                        save_optimizer_checkpoint(
                            model, optimizer, rank, train_config, epoch=epoch
                        )
                        print(" Saving the FSDP model checkpoints and optimizer using FULL_STATE_DICT")
                        print("=====================================================")                     
                if train_config.enable_fsdp:
                    dist.barrier()
            checkpoint_end_time = time.perf_counter() - checkpoint_start_time
            checkpoint_times.append(checkpoint_end_time)
            if eval_epoch_loss < best_val_loss:
                best_val_loss = eval_epoch_loss
                if train_config.enable_fsdp:
                    if rank==0:
                        print(f"best eval loss on epoch {epoch+1} is {best_val_loss}")
                else:
                    print(f"best eval loss on epoch {epoch+1} is {best_val_loss}")
            val_loss.append(best_val_loss)
            val_prep.append(eval_ppl)
        if train_config.enable_fsdp:
            if rank==0:
                print(f"Epoch {epoch+1}: train_perplexity={train_perplexity:.4f}, train_epoch_loss={train_epoch_loss:.4f}, epoch time {epoch_end_time}s")
        else:
            print(f"Epoch {epoch+1}: train_perplexity={train_perplexity:.4f}, train_epoch_loss={train_epoch_loss:.4f}, epoch time {epoch_end_time}s")
    avg_epoch_time = sum(epoch_times)/ len(epoch_times)
    avg_checkpoint_time = sum(checkpoint_times)/ len(checkpoint_times) if len(checkpoint_times) > 0 else 0
    avg_train_prep = sum(train_prep)/len(train_prep)
    avg_train_loss = sum(train_loss)/len(train_loss)
    if train_config.run_validation:
        avg_eval_prep = sum(val_prep)/len(val_prep) 
        avg_eval_loss = sum(val_loss)/len(val_loss) 

    results['avg_train_prep'] = avg_train_prep
    results['avg_train_loss'] = avg_train_loss
    if train_config.run_validation:
        results['avg_eval_prep'] = avg_eval_prep
        results['avg_eval_loss'] = avg_eval_loss
    results["avg_epoch_time"] = avg_epoch_time
    results["avg_checkpoint_time"] = avg_checkpoint_time
    if train_config.flop_counter:
        results["model_flops"]= TFlops
        
    
    #saving the training params including fsdp setting for reference.
    if train_config.enable_fsdp and not train_config.use_peft:
        save_train_params(train_config, fsdp_config, rank)
        
    return results

def evaluation(model,train_config, eval_dataloader, local_rank, tokenizer):
    """
    Evaluates the model on the given dataloader
    
    Args:
        model: The model to evaluate
        eval_dataloader: The dataloader containing the evaluation data
        local_rank: The rank of the current node in a distributed setting
        tokenizer: The tokenizer used to decode predictions
    
    Returns: eval_ppl, eval_epoch_loss
    """
    if train_config.enable_fsdp:
        world_size = int(os.environ["WORLD_SIZE"]) 
    model.eval()
    eval_preds = []
    eval_loss = 0.0  # Initialize evaluation loss
    with MemoryTrace() as memtrace:
        for step, batch in enumerate(tqdm(eval_dataloader,colour="green", desc="evaluating Epoch", dynamic_ncols=True)):
            if step > 5:
                break
            gc.collect(1)
            for key in batch.keys():
                if train_config.enable_fsdp:
                    batch[key] = batch[key].to(local_rank)
                else:
                    batch[key] = batch[key].to('cuda:0')
            # Ensure no gradients are computed for this scope to save memory
            with torch.no_grad():
                # Forward pass and compute loss
                outputs = model(**batch)
                loss = outputs.loss
                eval_loss += loss.detach().float()
            # Decode predictions and add to evaluation predictions list
            preds = torch.argmax(outputs.logits, -1)
            eval_preds.extend(
                tokenizer.batch_decode(preds.detach().cpu().numpy(), skip_special_tokens=True)
            )
    
    # If there's more than one CUDA device, reduce evaluation loss across all devices
    if torch.cuda.device_count() > 1 and train_config.enable_fsdp:
        dist.all_reduce(eval_loss, op=dist.ReduceOp.SUM)
    
    # Compute average loss and perplexity
    eval_epoch_loss = eval_loss / len(eval_dataloader)
    if train_config.enable_fsdp:
        eval_epoch_loss = eval_epoch_loss/world_size
    eval_ppl = torch.exp(eval_epoch_loss)
    
    # Print evaluation metrics
    if train_config.enable_fsdp:
        if local_rank==0:
            print(f" {eval_ppl=} {eval_epoch_loss=}")
    else:
        print(f" {eval_ppl=} {eval_epoch_loss=}")
        
    return eval_ppl, eval_epoch_loss

def freeze_transformer_layers(model, num_layer):
   for i, layer in enumerate(model.model.layers):
            if i < num_layer:
                for param in layer.parameters():
                    param.requires_grad = False


def check_frozen_layers_peft_model(model):
     for i, layer in enumerate(model.base_model.model.model.layers):
            for name, param in layer.named_parameters():
                print(f"Layer {i}, parameter {name}: requires_grad = {param.requires_grad}")
                
                
def setup():
    """Initialize the process group for distributed training"""
    dist.init_process_group("nccl")


def setup_environ_flags(rank):
    """Set environment flags for debugging purposes"""
    os.environ["TORCH_SHOW_CPP_STACKTRACES"] = str(1)
    os.environ["NCCL_ASYNC_ERROR_HANDLING"] = str(1)
    # os.environ["TORCH_DISTRIBUTED_DEBUG"] = "DETAIL"
    # This flag will help with CUDA memory fragmentations that can lead into OOM in some cases.
    # Note this is only availble in PyTorch Nighlies (as of July 30 2023)
    # os.environ['PYTORCH_CUDA_ALLOC_CONF']='expandable_segments:True' 
    if rank == 0:
        print(f"--> Running with torch dist debug set to detail")


def cleanup():
    """Clean up the process group after training"""
    dist.destroy_process_group()


def clear_gpu_cache(rank=None):
    """Clear the GPU cache for all ranks"""
    if rank == 0:
        print(f"Clearing GPU cache for all ranks")
    torch.cuda.empty_cache()


def get_parameter_dtypes(model):
    """Get the data types of model parameters"""
    parameter_dtypes = {}
    for name, parameter in model.named_parameters():
        parameter_dtypes[name] = parameter.dtype
    return parameter_dtypes

def print_model_size(model, config, rank: int = 0) -> None:
    """
    Print model name, the number of trainable parameters and initialization time.

    Args:
        model: The PyTorch model.
        model_name (str): Name of the model.
        init_time_start (float): Initialization start time.
        init_time_end (float): Initialization end time.
        rank (int, optional): Current process's rank. Defaults to 0.
    """
    if rank == 0:
        print(f"--> Model {config.model_name}")
        total_params = sum(p.numel() for p in model.parameters() if p.requires_grad)
        print(f"\n--> {config.model_name} has {total_params / 1e6} Million params\n")




def get_policies(cfg, rank):
    """Get the policies for mixed precision and fsdp wrapping"""
    
    verify_bfloat_support = (
    torch.version.cuda
    and torch.cuda.is_bf16_supported()
    and packaging.version.parse(torch.version.cuda).release >= (11, 0)
    and dist.is_nccl_available()
    and nccl.version() >= (2, 10)
    )


    mixed_precision_policy = None
    wrapping_policy = None

    # Mixed precision
    if cfg.mixed_precision:
        bf16_ready = verify_bfloat_support

        if bf16_ready and not cfg.use_fp16:
            mixed_precision_policy = bfSixteen_mixed
            if rank == 0:
                print(f"bFloat16 enabled for mixed precision - using bfSixteen policy")
        elif cfg.use_fp16:
            mixed_precision_policy = fpSixteen
            if rank == 0:
                print(f"FP16 enabled")
        else:
            print(f"bFloat16 support not present. Using FP32, and not mixed precision")
    wrapping_policy = get_llama_wrapper()
    return mixed_precision_policy, wrapping_policy

def save_train_params(train_config, fsdp_config, rank):
    """
    This function saves the train_config and FSDP config into a train_params.yaml.
    This will be used by converter script in the inference folder to fetch the HF model name or path.
    It also would be hepful as a log for future references.
    """
    # Convert the train_config and fsdp_config objects to dictionaries, 
    # converting all values to strings to ensure they can be serialized into a YAML file
    train_config_dict = {k: str(v) for k, v in vars(train_config).items() if not k.startswith('__')}
    fsdp_config_dict = {k: str(v) for k, v in vars(fsdp_config).items() if not k.startswith('__')}
    # Merge the two dictionaries into one
    train_params_dict = {**train_config_dict, **fsdp_config_dict}
    # Construct the folder name (follwoing FSDP checkpointing style) using properties of the train_config object
    folder_name = (
    train_config.dist_checkpoint_root_folder
    + "/"
    + train_config.dist_checkpoint_folder
    + "-"
    + train_config.model_name
    )

    save_dir = Path.cwd() / folder_name
    # If the directory does not exist, create it
    if not os.path.exists(save_dir):
        os.makedirs(save_dir)
    # Convert the dictionary to a YAML string
    config_yaml = yaml.dump(train_params_dict, indent=4)
    file_name = os.path.join(save_dir,'train_params.yaml')

    # Check if there's a directory with the same name as the file
    if os.path.isdir(file_name):
        print(f"Error: {file_name} is a directory, not a file.")
    else:
        # Write the YAML string to the file
        with open(file_name, 'w') as f:
            f.write(config_yaml)
        if rank==0:
            print(f"training params are saved in {file_name}")