apps/main/lingua_train.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 gc
import logging
import os
import sys
from contextlib import ExitStack
from copy import deepcopy
from dataclasses import asdict, dataclass, field
from pathlib import Path
from timeit import default_timer as timer
from typing import Any, Dict, Optional

import torch
import torch.distributed
import wandb
import xformers.profiler
from lingua.args import dataclass_from_dict, dump_config, flatten_dict
from lingua.data import (
    DataArgs,
    PackTokensState,
    build_dataloader_from_args,
    init_dataloader_state_from_args,
)
from lingua.tokenizers.build_tokenizer import TokenizerArgs
from omegaconf import OmegaConf
from pydantic import BaseModel
from torch.distributed._tensor import DTensor
from torch.distributed.checkpoint.stateful import Stateful
from torch.optim import lr_scheduler

from bytelatent.checkpoint import (
    CheckpointArgs,
    CheckpointManager,
    load_from_checkpoint,
)
from bytelatent.distributed import (
    DistributedArgs,
    EnvironmentArgs,
    check_model_value_range,
    clean_env,
    dist_mean_dict,
    get_device_mesh,
    get_is_master,
    get_world_size,
    init_signal_handler,
    parallelize_model,
    requeue_slurm_job,
    setup_env,
    setup_torch_distributed,
)
from bytelatent.logger import init_logger
from bytelatent.metrics import (
    GPUMemoryMonitor,
    LoggingArgs,
    MetricLogger,
    get_num_params,
)
from bytelatent.optim import OptimArgs, build_optimizer
from bytelatent.probe import AutoProbeD
from bytelatent.profiling import ProfilerArgs, maybe_run_profiler
from bytelatent.stool import StoolArgs, launch_job
from bytelatent.transformer import (
    LMTransformer,
    LMTransformerArgs,
    build_fsdp_grouping_plan,
    get_no_recompute_ops,
    get_num_flop_per_token,
    tp_parallelize,
)

logger = logging.getLogger()


class TrainArgs(BaseModel):
    name: str = "lingua"
    dump_dir: str = ""

    seed: int = 42

    # Number of gradient accumulation steps
    # Total batch size is batch_size*grad_acc_steps
    grad_acc_steps: int = 1

    gc_collect_freq: int = 1000
    probe_freq: int | None = None

    # Nb optimizer steps to take
    steps: int = 1000

    data: DataArgs
    optim: OptimArgs
    model: LMTransformerArgs
    distributed: DistributedArgs
    env: EnvironmentArgs

    checkpoint: CheckpointArgs
    profiling: ProfilerArgs
    logging: LoggingArgs

    # If set to None, eval is run locally otherwise it launches a new job with the given number of gpus
    async_eval_gpus: int | None = None
    eval: Any | None = None


@dataclass
class TrainState(Stateful):
    step: int  # Nb of steps taken by the optimizer
    acc_step: int  # Nb of accumulation steps done since last optimizer step
    scheduler: lr_scheduler.LambdaLR
    data_loader_state: PackTokensState

    def state_dict(self) -> Dict[str, Any]:
        return {
            "step": self.step,
            "acc_step": self.acc_step,
            "data_loader_state": self.data_loader_state,
            "scheduler": self.scheduler.state_dict(),
        }

    def load_state_dict(self, state_dict):
        self.step = state_dict["step"]
        self.acc_step = state_dict["acc_step"]
        self.data_loader_state = PackTokensState(**state_dict["data_loader_state"])
        self.scheduler.load_state_dict(state_dict["scheduler"])


def validate_train_args(args: TrainArgs, output_size: int):
    if args.model.vocab_size < 0:
        logger.info(f"Setting model output size to {args.model.vocab_size}")
        args.model.vocab_size = output_size
    assert (
        args.model.vocab_size == output_size
    ), "Vocab size should be the same as output size"

    assert args.dump_dir, "Dump dir not set"

    if args.checkpoint.path is None:
        logger.info(f"Setting checkpoint path to {args.checkpoint.path}")
        args.checkpoint.path = str(Path(args.dump_dir) / "checkpoints")

    for source in args.data.sources:
        data_path = os.path.join(args.data.root_dir, source)
        assert os.path.exists(data_path), f"{data_path} doesn't exist"

    if (
        args.distributed.dp_replicate
        * args.distributed.dp_shard
        * args.distributed.tp_size
        != get_world_size()
    ):
        assert get_world_size() % args.distributed.dp_shard == 0
        args.distributed.dp_replicate = get_world_size() // args.distributed.dp_shard

        assert args.distributed.dp_replicate % args.distributed.tp_size == 0
        args.distributed.dp_replicate = (
            args.distributed.dp_replicate // args.distributed.tp_size
        )

        logger.warning(
            f"Setting Data Parallel size to {args.distributed.dp_replicate * args.distributed.dp_shard}"
        )
        assert (
            args.distributed.dp_replicate
            * args.distributed.dp_shard
            * args.distributed.tp_size
            == get_world_size()
        )

        if args.distributed.fsdp_type == "no_shard":
            assert (
                args.distributed.dp_shard == 1
                and args.distributed.dp_replicate == get_world_size()
            )

    args.model.max_seqlen = args.data.seq_len

    if args.distributed.tp_size == 1:
        logger.warning(
            "Tensor parallelism has not been tested for a while, use at your own risk"
        )

    assert (
        args.probe_freq != args.profiling.mem_steps
    ), "Don't profile during probe step"
    assert (
        args.probe_freq != args.profiling.profile_steps
    ), "Don't profile during probe step"
    if args.logging.wandb is not None:
        args.logging.wandb.name = args.name

    if args.probe_freq is not None:
        assert (
            args.distributed.tp_size == 1
        ), "Probing not supported with tensor parallelism"
        assert (
            args.distributed.selective_activation_checkpointing is False
        ), "Probing not supported with selective activation checkpointing"


preemption_flag = dict(flag=False)


def set_preemption_flag(signum, frame):
    logger.warning("Signal handler called with signal " + str(signum))
    logger.warning("Preemption ! checkpointing asap and exiting.")
    preemption_flag["flag"] = True


def every_n_steps(train_state, freq, acc_step=None, acc_freq=None):
    test = train_state.step % freq == 0
    if acc_step is not None:
        test = test and (train_state.acc_step == acc_step)
    elif acc_freq is not None:
        test = test and ((train_state.acc_step % acc_freq) == 0)
    return test


def train(args: TrainArgs):
    with ExitStack() as context_stack:
        tokenizer_args = TokenizerArgs(
            name=args.data.name,
            init_kwargs=args.data.tokenizer.init_kwargs,
        )
        tokenizer = tokenizer_args.build()
        validate_train_args(
            args,
            tokenizer.n_words,
        )
        if get_is_master():
            os.makedirs(args.dump_dir, exist_ok=True)
            dump_config(args, Path(args.dump_dir) / "config.yaml")
        init_logger(Path(args.dump_dir) / "train.log")
        init_signal_handler(set_preemption_flag)  # For handling preemption signals.
        setup_env(args.env)
        setup_torch_distributed(args.distributed)
        world_mesh = get_device_mesh(args.distributed)
        logger.info(f"Starting job: {args.name}")

        # build dataloader
        # need dp world size and rank
        dp_mesh = world_mesh["dp_replicate"]
        dp_degree = dp_mesh.size()
        dp_rank = dp_mesh.get_local_rank()
        if args.distributed.dp_shard > 1:
            dp_rank = dp_rank * dp_degree + world_mesh["dp_shard"].get_local_rank()
            dp_degree *= world_mesh["dp_shard"].size()

        logger.info(f"Running on dp rank : {dp_rank}")
        logger.info(f"Running on dp size : {dp_degree}")

        torch.manual_seed(args.seed)
        logger.info("Building model")

        # Initializing Model in meta device allows us to initialize models much bigger than 1 gpu's memory
        with torch.device("meta"):
            model = LMTransformer(args.model)
        logger.info("Model is built !")

        model_param_count = get_num_params(model)

        model = parallelize_model(
            model,
            world_mesh,
            args.model,
            args.distributed,
            fsdp_grouping_plan=build_fsdp_grouping_plan(args.model),
            tp_parallelize=tp_parallelize,
            no_recompute_ops=get_no_recompute_ops(),
        )

        # Once we shard the model on different gpus we can actually initialize the model
        # First we create empty tensors of the correct shapes
        model = model.to_empty(device="cuda")
        # Then we init the model. Please make sure this function initializes *ALL* parameters
        # and buffers, otherwise you will have random values in the unitialized tensors
        # which will silently fail (give nan gradients for example)

        if args.checkpoint.init_ckpt_path:
            logger.info(f"Loading initial model from {args.checkpoint.init_ckpt_path}")
            load_from_checkpoint(
                args.checkpoint.init_ckpt_path, model, model_key="model"
            )  # Put model_key="" if its directly the model checkpoint
            model.rope_embeddings.reset_parameters()  # For RoPe initialization since it's a buffer it might not be loaded
        else:
            with torch.random.fork_rng(devices=[torch.cuda.current_device()]):
                torch.manual_seed(args.model.seed)
                model.init_weights()
        check_model_value_range(model, range=10.0, std=1.0)

        # log model size

        logger.info(f"Model size: {model_param_count:,} total parameters")

        gpu_memory_monitor = GPUMemoryMonitor("cuda")
        logger.info(
            f"GPU capacity: {gpu_memory_monitor.device_name} ({gpu_memory_monitor.device_index}) "
            f"with {gpu_memory_monitor.device_capacity_gib:.2f}GiB memory"
        )
        logger.info(f"GPU memory usage: {gpu_memory_monitor}")

        # build optimizer after apply parallelisms to the model
        optimizer, scheduler = build_optimizer(model, args.optim, args.steps)
        data_loader_state = init_dataloader_state_from_args(
            args.data, dp_rank, dp_degree
        )

        train_state = TrainState(
            step=0,
            acc_step=0,
            data_loader_state=data_loader_state,
            scheduler=scheduler,
        )

        checkpoint = CheckpointManager.instantiate_and_make_dir(args.checkpoint)
        checkpoint.load(model, optimizer, train_state, world_mesh)
        # Either load from latest checkpoint or start from scratch
        if args.probe_freq is not None:
            if get_is_master():
                os.makedirs(Path(args.dump_dir) / "probe", exist_ok=True)
            torch.distributed.barrier()
            probe = AutoProbeD(
                model,
                (
                    Path(args.dump_dir) / "probe" / f"probe.{dp_rank}.jsonl"
                    if (dp_rank % 128 == 0)
                    else None
                ),
            )
            probe_mod = model._orig_mod if args.distributed.compile else model

        gc.disable()

        # train loop
        model.train()
        metric_logger = context_stack.enter_context(
            MetricLogger(Path(args.dump_dir) / "metrics.jsonl", args)
        )
        data_loader = context_stack.enter_context(
            build_dataloader_from_args(
                args.data,
                state=train_state.data_loader_state,
            )
        )
        torch_profiler = context_stack.enter_context(
            maybe_run_profiler(args.dump_dir, model, args.profiling)
        )

        nwords_since_last_log = 0
        time_last_log = timer()
        gc.collect()
        while train_state.step < args.steps:
            # We constrain train_state.acc_step to be in range 0 to args.grad_acc_steps - 1
            train_state.acc_step += 1
            train_state.acc_step = train_state.acc_step % args.grad_acc_steps

            # get batch
            curr_lr = float(optimizer.param_groups[0]["lr"])
            data_load_start = timer()
            batch, train_state.data_loader_state = next(data_loader)
            batch = torch.tensor(
                batch,
                dtype=torch.long,
            )

            if every_n_steps(train_state, args.gc_collect_freq, acc_step=0):
                logger.info("garbage collection")
                # we do garbage collection manually otherwise different processes
                # run the GC at different times so they slow down the whole pipeline
                gc.collect()

            input_ids = batch[:, :, 0].cuda()
            labels = batch[:, :, 1].cuda()
            data_load_time = round(timer() - data_load_start, 4)
            nwords_since_last_log += input_ids.numel()

            bsz, seqlen = labels.shape

            # forward
            start_timer = torch.cuda.Event(enable_timing=True)
            end_timer = torch.cuda.Event(enable_timing=True)
            start_timer.record()

            # This is an automatic probe that will compute statistics
            # of all linears' inputs, weights and outputs
            # along with attention logits and entropy
            # both in forward and backward pass
            if (args.probe_freq is not None) and every_n_steps(
                train_state, args.probe_freq, acc_step=1 % args.grad_acc_steps
            ):
                # Here we do a fake forward and backward pass on a smaller
                # batch size to avoid OOM
                # This assumes the model has no stateful layers (batch norm..)
                assert (
                    next(probe_mod.parameters()).grad is None
                ), "Can't probe model if grads are not reset"

                with probe:
                    probe.metadata = {
                        "it": train_state.step,
                        "global_step": train_state.step,
                        "loop": "lingua",
                    }
                    # Non compiled model uses roughly 2x memory in our exps
                    # So we divide bsz by 2 or seqlen by 2
                    probe_bsz = max(1, bsz // 2)
                    probe_seq = seqlen if (bsz // 2 >= 1) else (seqlen // 2)
                    probe_loss = probe_mod(
                        input_ids[:probe_bsz, :probe_seq],
                        labels[:probe_bsz, :probe_seq],
                    )
                    probe_loss.backward()
                    # We zero grads to cancel this fake step
                    optimizer.zero_grad()

                assert (
                    next(probe_mod.parameters()).grad is None
                ), "Probe model shouldn't have grads at this point"
            loss = model(input_ids, labels)

            # We scale loss with grad_acc_steps so the gradient is the same
            # regardless of grad_acc_steps
            loss = loss / args.grad_acc_steps
            # backward on scaled loss to create scaled gradients
            loss.backward()
            # For logging we undo that scaling
            loss = loss.detach() * args.grad_acc_steps

            grad_norm = torch.nn.utils.clip_grad_norm_(
                model.parameters(), max_norm=args.optim.clip, foreach=True
            )

            grad_norm = (
                grad_norm.full_tensor() if isinstance(grad_norm, DTensor) else grad_norm
            ).item()

            # optimizer step
            if train_state.acc_step == 0:
                optimizer.step()
                scheduler.step()
                optimizer.zero_grad()
                train_state.step += 1

            # updates the scale for next iteration
            # training iteration complete
            end_timer.record()

            torch.cuda.synchronize()

            curr_iter_time = round(start_timer.elapsed_time(end_timer) * 1e-3, 4)

            # if profiler is active
            if torch_profiler:
                xformers.profiler.step()

            # log metrics
            if every_n_steps(
                train_state,
                args.logging.freq,
                acc_step=None if args.logging.acc_freq else 0,
                acc_freq=args.logging.acc_freq,
            ):
                time_delta = timer() - time_last_log
                wps = nwords_since_last_log / (time_delta * args.distributed.tp_size)

                gpu_mem_stats = gpu_memory_monitor.get_peak_stats()

                total_acc_steps = (
                    args.grad_acc_steps * train_state.step + train_state.acc_step
                )
                tokens_per_gpu = (
                    total_acc_steps * args.data.batch_size * args.data.seq_len
                )
                total_tokens = dp_degree * tokens_per_gpu
                # This is an estimate and the correct values may change
                # if you change the architecture
                # Use xformer's analyze profile trace to get actual measurement
                FLOPS = (
                    get_num_flop_per_token(
                        model_param_count - args.model.vocab_size * args.model.dim,
                        args.model.n_layers,
                        args.model.dim,
                        args.data.seq_len,
                    )
                    * wps
                )
                metrics = flatten_dict(
                    {
                        "global_step": train_state.step,
                        "acc_step": train_state.acc_step,
                        "speed": {
                            "wps": wps,
                            "FLOPS": FLOPS,
                            "curr_iter_time": curr_iter_time,
                            "data_load_time": data_load_time,
                        },
                        "optim": {
                            "grad_norm": grad_norm,
                            "lr": curr_lr,
                            "total_tokens": total_tokens,
                        },
                        "memory": gpu_mem_stats._asdict(),
                    },
                    sep="/",
                )

                to_sync = {}
                to_sync["loss/out"] = loss.item()
                metrics.update(dist_mean_dict(to_sync))

                if get_is_master():
                    metric_logger.log(metrics)

                gpu_memory_monitor.reset_peak_stats()
                nwords_since_last_log = 0
                time_last_log = timer()
                logger.info(
                    f"step: {train_state.step}"
                    f"  acc: {train_state.acc_step}"
                    f"  loss: {round(loss.item(),4):>7}"
                    f"  grad: {grad_norm:.2e}"
                    f"  flops: {FLOPS:.2e}"
                    f"  wps: {wps:.2e}"
                    f"  iter: {curr_iter_time:>7}"
                    f"  data: {data_load_time:>5}"
                    f"  lr: {curr_lr:.2e}"
                    f"  mem: {gpu_mem_stats.max_active_pct:.0f}%"
                    f"  pow: {gpu_mem_stats.power_draw/1000} W"
                )

            saved = False
            if every_n_steps(
                train_state, args.checkpoint.dump.every, acc_step=0
            ) or every_n_steps(train_state, args.checkpoint.eval.every, acc_step=0):
                saved = checkpoint.save(
                    model,
                    optimizer,
                    train_state,
                    args,
                    device_mesh=world_mesh,
                )

            if args.eval is not None and every_n_steps(
                train_state, args.checkpoint.eval.every, acc_step=0
            ):
                from bytelatent.eval import EVAL_FOLDER_NAME, EvalArgs, launch_eval

                eval_args = dataclass_from_dict(EvalArgs, args.eval)

                eval_args.global_step = train_state.step
                eval_args.ckpt_dir = str(checkpoint.existing_saves[-1])
                eval_args.dump_dir = str(
                    os.path.join(
                        args.dump_dir,
                        "evals",
                        EVAL_FOLDER_NAME.format(train_state.step),
                    )
                )
                eval_args.metric_log_dir = args.dump_dir
                if args.async_eval_gpus is None:
                    launch_eval(eval_args)
                elif get_is_master():
                    if wandb.run is not None and args.logging.wandb is not None:
                        eval_args.wandb = deepcopy(args.logging.wandb)
                    assert args.async_eval_gpus > 0
                    logger.info(f"Launching evals on {args.async_eval_gpus} gpus")
                    with clean_env():
                        launch_job(
                            StoolArgs(
                                asdict(eval_args),
                                script="apps.main.eval",
                                copy_code=False,
                                nodes=args.async_eval_gpus // 8,
                                qos="lowest",
                            )
                        )

            if preemption_flag["flag"]:
                if not saved:
                    checkpoint.save(
                        model,
                        optimizer,
                        train_state,
                        args,
                        device_mesh=world_mesh,
                    )
                requeue_slurm_job()
                sys.exit(0)

    if not saved:
        checkpoint.save(
            model,
            optimizer,
            train_state,
            args,
            device_mesh=world_mesh,
        )
    gc.collect()


def main():
    """
    The command line interface here uses OmegaConf https://omegaconf.readthedocs.io/en/2.3_branch/usage.html#from-command-line-arguments
    This accepts arguments as a dot list
    So if the dataclass looks like

    @dataclass
    class DummyArgs:
        name: str
        model: LMTransformerArgsgs

    @dataclass
    class LMTransformerArgsgs:
        dim: int

    Then you can pass model.dim=32 to change values in LMTransformerArgsgs
    or just name=tictac for top level attributes.

    The behavior here is as follows:
    1. We instantiate TrainArgs with its default values
    2. We override those default values with the ones in the provided config file
    3. We override the result with the additional arguments provided through command line

    For example, if the config is the following

    model:
        dim: 128
        n_layers: 4

    and you call train.py with train.py model.dim=64

    Then the final TrainArgs will have

    model:
        dim: 64
        n_layers: 4

    Plus all the default values in TrainArgs dataclass.
    """
    cli_args = OmegaConf.from_cli()
    file_cfg = OmegaConf.load(cli_args.config)
    # We remove 'config' attribute from config as the underlying DataClass does not have it
    del cli_args.config

    default_cfg = OmegaConf.structured(TrainArgs())
    cfg = OmegaConf.merge(default_cfg, file_cfg, cli_args)
    cfg = OmegaConf.to_object(cfg)

    train(cfg)


if __name__ == "__main__":
    main()