train.py

# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the BSD-style license found in the
# LICENSE file in the root directory of this source tree.

import importlib
import os
import time
from datetime import timedelta
from typing import Any, Generator, Iterable, Optional

import torch
from torch.distributed.elastic.multiprocessing.errors import record

import torchtitan.protocols.train_spec as train_spec_module
from torchtitan.components.checkpoint import CheckpointManager
from torchtitan.components.dataloader import DataloaderExhaustedError
from torchtitan.components.ft import FTManager, maybe_semi_sync_training
from torchtitan.components.loss import rescale_accumulated_loss
from torchtitan.components.metrics import (
    build_metrics_processor,
    ensure_pp_loss_visible,
)
from torchtitan.config import ConfigManager, JobConfig, TORCH_DTYPE_MAP
from torchtitan.distributed import ParallelDims, utils as dist_utils
from torchtitan.protocols.model_converter import build_model_converters
from torchtitan.tools import utils
from torchtitan.tools.logging import init_logger, logger
from torchtitan.tools.profiling import (
    maybe_enable_memory_snapshot,
    maybe_enable_profiling,
)


class Trainer(torch.distributed.checkpoint.stateful.Stateful):
    # core configs
    job_config: JobConfig
    parallel_dims: ParallelDims
    train_spec: train_spec_module.TrainSpec

    # swappable training components in TrainSpec
    tokenizer: train_spec_module.BaseTokenizer | None
    dataloader: train_spec_module.BaseDataLoader
    model_parts: list[torch.nn.Module]
    loss_fn: train_spec_module.LossFunction
    optimizers: train_spec_module.OptimizersContainer
    lr_schedulers: train_spec_module.LRSchedulersContainer
    validator: train_spec_module.BaseValidator
    metrics_processor: train_spec_module.MetricsProcessor
    model_args: train_spec_module.BaseModelArgs

    # non-swappable training components
    checkpointer: CheckpointManager
    ft_manager: FTManager

    # runtime utilities
    device: torch.device
    gc_handler: utils.GarbageCollection
    train_context: Generator[None, None, None]
    gradient_accumulation_steps: int
    pp_has_first_stage: bool
    pp_has_last_stage: bool

    # additional training states
    step: int
    ntokens_seen: int

    # Enable debug tracing on failure: https://pytorch.org/docs/stable/elastic/errors.html
    @record
    def __init__(self, job_config: JobConfig):
        torch._C._log_api_usage_once("torchtitan.train")

        self.job_config = job_config

        logger.info(f"Starting job: {job_config.job.description}")

        if job_config.experimental.custom_import:
            importlib.import_module(job_config.experimental.custom_import)

        device_module, device_type = utils.device_module, utils.device_type
        self.device = torch.device(f"{device_type}:{int(os.environ['LOCAL_RANK'])}")
        # Device has to be set before creating TorchFT manager.
        device_module.set_device(self.device)

        # init distributed and build meshes
        dist_utils.init_distributed(
            job_config.comm,
            enable_cpu_backend=job_config.training.enable_cpu_offload,
            base_folder=job_config.job.dump_folder,
        )

        job_config.maybe_log()

        world_size = int(os.environ["WORLD_SIZE"])

        parallelism_config = job_config.parallelism
        self.parallel_dims = parallel_dims = self._create_parallel_dims(
            parallelism_config, world_size
        )

        world_mesh = parallel_dims.world_mesh
        if parallel_dims.dp_enabled:
            dp_mesh = world_mesh["dp"]
            dp_degree, dp_rank = dp_mesh.size(), dp_mesh.get_local_rank()
        else:
            dp_degree, dp_rank = 1, 0

        self.ft_manager = FTManager(job_config.fault_tolerance)
        dp_degree, dp_rank = self.ft_manager.get_dp_info(dp_degree, dp_rank)

        # take control of garbage collection to avoid stragglers
        self.gc_handler = utils.GarbageCollection(
            gc_freq=job_config.training.gc_freq, debug=job_config.training.gc_debug
        )

        # Set random seed, and maybe enable deterministic mode
        # (mainly for debugging, expect perf loss).
        dist_utils.set_determinism(
            world_mesh,
            self.device,
            job_config.training.seed,
            job_config.training.deterministic,
        )
        self.train_spec = train_spec_module.get_train_spec(job_config.model.name)

        # build tokenizer and dataloader
        self.tokenizer = (
            self.train_spec.build_tokenizer_fn(job_config)
            if self.train_spec.build_tokenizer_fn is not None
            else None
        )

        self.dataloader = self.train_spec.build_dataloader_fn(
            dp_world_size=dp_degree,
            dp_rank=dp_rank,
            tokenizer=self.tokenizer,
            job_config=job_config,
        )

        # build model (using meta init)
        model_args = self.train_spec.model_args[job_config.model.flavor]
        # set the model args from training job configs
        model_args.update_from_config(job_config)
        self.model_args = model_args

        logger.info(
            f"Building {job_config.model.name} {job_config.model.flavor} with {model_args}"
        )

        with (
            torch.device("meta"),
            utils.set_default_dtype(TORCH_DTYPE_MAP[job_config.training.dtype]),
        ):
            model = self.train_spec.model_cls(model_args)

        # Build the collection of model converters. No-op if `model.converters` empty
        model_converters = build_model_converters(job_config, parallel_dims)
        model_converters.convert(model)

        # metrics logging
        build_metrics_processor_fn = (
            build_metrics_processor
            if self.train_spec.build_metrics_processor_fn is None
            else self.train_spec.build_metrics_processor_fn
        )
        self.metrics_processor = build_metrics_processor_fn(
            job_config, parallel_dims, model_args
        )
        color = self.metrics_processor.color

        # calculate model size and flops per token
        (
            model_param_count,
            self.metrics_processor.num_flops_per_token,
        ) = model_args.get_nparams_and_flops(model, job_config.training.seq_len)

        logger.info(
            f"{color.blue}Model {job_config.model.name} {job_config.model.flavor} "
            f"{color.red}size: {model_param_count:,} total parameters{color.reset}"
        )

        # move sharded model to CPU/GPU and initialize weights via DTensor
        if job_config.checkpoint.create_seed_checkpoint:
            init_device = "cpu"
            buffer_device = None
        elif job_config.training.enable_cpu_offload:
            init_device = "cpu"
            buffer_device = device_type
        else:
            init_device = device_type
            buffer_device = None

        self.loss_fn = self.train_spec.build_loss_fn(
            job_config, parallel_dims=parallel_dims, ft_manager=self.ft_manager
        )

        # verify batch sizes
        global_batch_size = job_config.training.global_batch_size
        if global_batch_size < 0:
            # This global batch size results in 1 gradient accumulation
            # step.
            global_batch_size = job_config.training.local_batch_size * dp_degree
        assert global_batch_size > 0
        assert (
            global_batch_size % (job_config.training.local_batch_size * dp_degree) == 0
        ), (
            f"global batch size must be multiple of local batch size times "
            f"data-parallel degree ({global_batch_size} "
            f"% ({job_config.training.local_batch_size} * {dp_degree}) != 0)"
        )

        # calculate gradient accumulation steps
        self.gradient_accumulation_steps = global_batch_size // (
            job_config.training.local_batch_size * dp_degree
        )
        assert self.gradient_accumulation_steps > 0
        self.loss_fn = rescale_accumulated_loss(
            self.loss_fn, self.gradient_accumulation_steps
        )

        # apply parallelisms and initialization
        if parallel_dims.pp_enabled:
            if not self.train_spec.pipelining_fn:
                raise RuntimeError(
                    f"Pipeline Parallel is enabled but {job_config.model.name} "
                    f"does not support pipelining"
                )

            # apply both PT-D Pipeline Parallel and SPMD-style PT-D techniques
            (
                self.pp_schedule,
                self.model_parts,
                self.pp_has_first_stage,
                self.pp_has_last_stage,
            ) = self.train_spec.pipelining_fn(
                model,
                parallel_dims,
                job_config,
                self.device,
                model_args,
                self.train_spec.parallelize_fn,
                self.loss_fn,
            )
            # when PP is enabled, `model` obj is no longer used after this point,
            # model_parts is used instead
            del model

            for m in self.model_parts:
                m.to_empty(device=init_device)
                with torch.no_grad():
                    m.init_weights(buffer_device=buffer_device)
                m.train()

            # confirm that user will be able to view loss metrics on the console
            ensure_pp_loss_visible(parallel_dims, job_config, color)
        else:
            # apply PT-D Tensor Parallel, activation checkpointing, torch.compile, Data Parallel
            model = self.train_spec.parallelize_fn(model, parallel_dims, job_config)

            model.to_empty(device=init_device)
            with torch.no_grad():
                model.init_weights(buffer_device=buffer_device)
            model.train()

            self.model_parts = [model]

        self.ft_manager.maybe_set_all_reduce_hook(self.model_parts)

        # initialize device memory monitor and get peak flops for MFU calculation
        device_memory_monitor = self.metrics_processor.device_memory_monitor
        gpu_peak_flops = utils.get_peak_flops(device_memory_monitor.device_name)
        logger.info(f"WORLD_SIZE: {world_size}")
        logger.info(f"NUM NODES: {world_size // 4}")
        logger.info(f"Peak FLOPS used for computing MFU: {gpu_peak_flops:.3e}")
        device_mem_stats = device_memory_monitor.get_peak_stats()
        logger.info(
            f"{device_type.upper()} memory usage for model: "
            f"{device_mem_stats.max_reserved_gib:.2f}GiB"
            f"({device_mem_stats.max_reserved_pct:.2f}%)"
        )

        # build optimizer after applying parallelisms to the model
        self.optimizers = self.train_spec.build_optimizers_fn(
            self.model_parts, job_config.optimizer, parallel_dims, self.ft_manager
        )
        self.lr_schedulers = self.train_spec.build_lr_schedulers_fn(
            self.optimizers, job_config.lr_scheduler, job_config.training.steps
        )
        # Post optimizer step model converters hook.
        # e.g. calculate float8 dynamic amax/scale for all-parameter for FSDP2
        # where it issues a single all-reduce for all parameters at once for better performance
        self.optimizers.register_step_post_hook(
            lambda *args, **kwargs: model_converters.post_optimizer_hook(
                self.model_parts
            )
        )
        self.metrics_processor.optimizers = self.optimizers
        self.metrics_processor.model_parts = self.model_parts

        # Initialize trainer states that will be saved in checkpoint.
        # These attributes must be initialized before checkpoint loading.
        self.step = 0
        self.ntokens_seen = 0

        self.checkpointer = CheckpointManager(
            dataloader=self.dataloader,
            model_parts=self.model_parts,
            optimizers=self.optimizers,
            lr_schedulers=self.lr_schedulers,
            states={"train_state": self},
            checkpoint_config=job_config.checkpoint,
            sd_adapter=(
                self.train_spec.state_dict_adapter(
                    model_args, job_config.model.hf_assets_path
                )
                if self.train_spec.state_dict_adapter
                else None
            ),
            base_folder=job_config.job.dump_folder,
            ft_manager=self.ft_manager,
        )

        loss_parallel_enabled = (
            parallel_dims.tp_enabled and not parallelism_config.disable_loss_parallel
        )
        self.train_context = dist_utils.get_train_context(
            loss_parallel_enabled,
            parallelism_config.enable_compiled_autograd,
        )
        self.maybe_enable_amp = dist_utils.maybe_enable_amp(
            parallel_dims,
            job_config.training.mixed_precision_param,
            device_type,
        )

        # Build validator if validation is configured
        if job_config.validation.enable:
            assert self.train_spec.build_validator_fn is not None

            pp_schedule, pp_has_first_stage, pp_has_last_stage = (
                (
                    self.pp_schedule,
                    self.pp_has_first_stage,
                    self.pp_has_last_stage,
                )
                if parallel_dims.pp_enabled
                else (None, None, None)
            )

            self.validator = self.train_spec.build_validator_fn(
                job_config=job_config,
                dp_world_size=dp_degree,
                dp_rank=dp_rank,
                tokenizer=self.tokenizer,
                parallel_dims=parallel_dims,
                loss_fn=self.loss_fn,
                validation_context=self.train_context,
                maybe_enable_amp=self.maybe_enable_amp,
                metrics_processor=self.metrics_processor,
                pp_schedule=pp_schedule,
                pp_has_first_stage=pp_has_first_stage,
                pp_has_last_stage=pp_has_last_stage,
            )

        logger.info(
            "Trainer is initialized with "
            f"local batch size {job_config.training.local_batch_size}, "
            f"global batch size {global_batch_size}, "
            f"gradient accumulation steps {self.gradient_accumulation_steps}, "
            f"sequence length {job_config.training.seq_len}, "
            f"total steps {job_config.training.steps} "
            f"(warmup {job_config.lr_scheduler.warmup_steps})"
        )

    def _create_parallel_dims(self, parallelism_config, world_size) -> ParallelDims:
        return ParallelDims(
            dp_shard=parallelism_config.data_parallel_shard_degree,
            dp_replicate=parallelism_config.data_parallel_replicate_degree,
            cp=parallelism_config.context_parallel_degree,
            tp=parallelism_config.tensor_parallel_degree,
            pp=parallelism_config.pipeline_parallel_degree,
            ep=parallelism_config.expert_parallel_degree,
            etp=parallelism_config.expert_tensor_parallel_degree,
            world_size=world_size,
        )

    def batch_generator(
        self, data_iterable: Iterable[tuple[dict[str, torch.Tensor], torch.Tensor]]
    ) -> Iterable[tuple[dict[str, torch.Tensor], torch.Tensor]]:
        """Returns an iterator that processes batches from the data iterator."""
        device_type = utils.device_type
        data_iterator = iter(data_iterable)

        while True:
            data_load_start = time.perf_counter()
            try:
                batch = next(data_iterator)
            except StopIteration as ex:
                # If data runs out during gradient accumulation, that
                # entire step will not be executed.
                raise DataloaderExhaustedError() from ex
            input_dict, labels = batch
            ntokens_batch = labels.numel()
            self.ntokens_seen += ntokens_batch
            self.metrics_processor.ntokens_since_last_log += ntokens_batch
            self.metrics_processor.data_loading_times.append(
                time.perf_counter() - data_load_start
            )

            # Move tensors to the appropriate device
            for k, v in input_dict.items():
                if isinstance(v, torch.Tensor):
                    input_dict[k] = v.to(device_type)
            labels = labels.to(device_type)

            yield input_dict, labels

    def forward_backward_step(
        self, input_dict: dict[str, torch.Tensor], labels: torch.Tensor
    ) -> torch.Tensor:
        model_parts = self.model_parts
        parallel_dims = self.parallel_dims

        inputs = input_dict["input"]
        extra_inputs = {k: v for k, v in input_dict.items() if k != "input"}
        # For arguments, like attention_masks, we have to put them in a separate
        # dict as extra_inputs are not forwarded to other stages in PP, but
        # extra_kwargs are.
        extra_kwargs = {}

        if getattr(self.model_args, "use_flex_attn", False):
            extra_kwargs["attention_masks"] = model_parts[0].get_attention_masks(
                input_batch=inputs,
                tokenizer=self.tokenizer,
                extra_inputs=extra_inputs,
            )

        # apply context parallelism if cp is enabled
        # ensure CP handles the separate freqs_cis buffer for each pp stage
        inputs = input_dict["input"]
        extra_inputs = {k: v for k, v in input_dict.items() if k != "input"}
        cp_mesh = parallel_dims.world_mesh["cp"] if parallel_dims.cp_enabled else None

        optional_context_parallel_ctx = (
            dist_utils.create_context_parallel_ctx(
                cp_mesh=parallel_dims.world_mesh["cp"],
                cp_buffers=[inputs, labels] + [m.freqs_cis for m in model_parts],
                cp_seq_dims=[1, 1] + [0 for _ in model_parts],
                cp_no_restore_buffers={inputs, labels},
                cp_rotate_method=self.job_config.parallelism.context_parallel_rotate_method,
            )
            if parallel_dims.cp_enabled
            else None
        )

        if parallel_dims.pp_enabled:
            # Pipeline Parallel forward / backward inside step() call
            with self.train_context(optional_context_parallel_ctx):
                targets, losses = (
                    (labels, []) if self.pp_has_last_stage else (None, None)
                )
                if self.pp_has_first_stage:
                    self.pp_schedule.step(
                        inputs,
                        **extra_inputs,
                        **extra_kwargs,
                        target=targets,
                        losses=losses,
                        return_outputs=False,
                    )
                else:
                    self.pp_schedule.step(
                        **extra_kwargs,
                        target=targets,
                        losses=losses,
                        return_outputs=False,
                    )

            # accumulate losses across pipeline microbatches
            # TODO: PP+FSDP unexpectedly puts the loss back to the CPU
            loss = (
                # using sum instead of mean because we already rescale the
                # loss_fn down by a factor of n_microbatches in
                # torchtitan/distributed/pipeline_parallel.py
                torch.sum(torch.stack(losses)).to(self.device)
                if self.pp_has_last_stage
                else torch.tensor([-1.0], device=self.device)
            )
        else:
            # Non-PP forward / backward
            with self.train_context(optional_context_parallel_ctx):
                assert len(model_parts) == 1
                with self.maybe_enable_amp:
                    pred = model_parts[0](inputs, **extra_inputs, **extra_kwargs)
                    loss = self.loss_fn(pred, labels)
                # need to free pred before bwd to avoid peaking memory
                del pred
                loss.backward()

        return loss

    def train_step(
        self, data_iterator: Iterable[tuple[dict[str, torch.Tensor], torch.Tensor]]
    ):
        self.optimizers.zero_grad()
        # Save the current step learning rate for logging
        lr = self.lr_schedulers.schedulers[0].get_last_lr()[0]

        # Keep these variables local to shorten the code as these are
        # the major variables that are used in the training loop.
        parallel_dims = self.parallel_dims

        accumulated_losses = []
        # If data runs out during gradient accumulation, that
        # entire step will not be executed.
        for _microbatch in range(self.gradient_accumulation_steps):
            input_dict, labels = next(data_iterator)
            loss = self.forward_backward_step(input_dict, labels)
            accumulated_losses.append(loss.detach())

        grad_norm = dist_utils.clip_grad_norm_(
            [p for m in self.model_parts for p in m.parameters()],
            self.job_config.training.max_norm,
            foreach=True,
            pp_mesh=(
                parallel_dims.world_mesh["pp"] if parallel_dims.pp_enabled else None
            ),
            ep_enabled=parallel_dims.ep_enabled,
        )
        self.checkpointer.maybe_wait_for_staging()
        self.optimizers.step()
        self.lr_schedulers.step()

        # Reduce the data collected over gradient accumulation steps.
        loss = torch.sum(torch.stack(accumulated_losses))

        # log metrics
        if not self.metrics_processor.should_log(self.step):
            return

        if parallel_dims.dp_cp_enabled:
            loss = loss.detach()
            ft_pg = self.ft_manager.loss_sync_pg
            global_avg_loss, global_max_loss, global_ntokens_seen = (
                dist_utils.dist_mean(loss, parallel_dims.world_mesh["dp_cp"], ft_pg),
                dist_utils.dist_max(loss, parallel_dims.world_mesh["dp_cp"], ft_pg),

                # we are getting the global tokens seen from a dist sum
                dist_utils.dist_sum(
                    torch.tensor(
                        self.ntokens_seen, dtype=torch.int64, device=self.device
                    ),
                    parallel_dims.world_mesh["dp_cp"],
                    ft_pg,
                ),
            )
        else:
            global_avg_loss = global_max_loss = loss.detach().item()
            global_ntokens_seen = self.ntokens_seen

        extra_metrics = {
            "n_tokens_seen": global_ntokens_seen,
            "lr": lr,
        }
        self.metrics_processor.log(
            self.step,
            global_avg_loss,
            global_max_loss,
            grad_norm.item(),
            extra_metrics=extra_metrics,
        )

    @record
    def train(self):
        job_config = self.job_config

        self.checkpointer.load(step=job_config.checkpoint.load_step)
        logger.info(f"Training starts at step {self.step + 1}")

        leaf_folder = (
            ""
            if not self.ft_manager.enabled
            else f"replica_{self.ft_manager.replica_id}"
        )
        with (
            maybe_enable_profiling(
                job_config.profiling,
                global_step=self.step,
                base_folder=job_config.job.dump_folder,
                leaf_folder=leaf_folder,
            ) as torch_profiler,
            maybe_enable_memory_snapshot(
                job_config.profiling,
                global_step=self.step,
                base_folder=job_config.job.dump_folder,
                leaf_folder=leaf_folder,
            ) as memory_profiler,
            maybe_semi_sync_training(
                job_config.fault_tolerance,
                ft_manager=self.ft_manager,
                model=self.model_parts[0],
                n_layers=(
                    self.model_args.n_layers
                    if hasattr(self.model_args, "n_layers")
                    else 0
                ),
                optimizer=self.optimizers,
                fragment_fn=(
                    self.train_spec.fragment_fn
                    if hasattr(self.train_spec, "fragment_fn")
                    else None
                ),
            ),
        ):
            data_iterator = self.batch_generator(self.dataloader)
            while self.should_continue_training():
                self.step += 1
                self.gc_handler.run(self.step)
                try:
                    self.train_step(data_iterator)
                except DataloaderExhaustedError:
                    logger.warning("Ran out of data; last step was canceled.")
                    break

                self.checkpointer.save(
                    self.step, last_step=(self.step == job_config.training.steps)
                )

                # Run validation if validator is available
                if (
                    self.job_config.validation.enable
                    and self.validator.should_validate(self.step)
                ):
                    with self.loss_fn.no_rescale():
                        self.validator.validate(self.model_parts, self.step)

                # signal the profiler that the next profiling step has started
                if torch_profiler:
                    torch_profiler.step()
                if memory_profiler:
                    memory_profiler.step()

                # reduce timeout after first train step for faster signal
                # (assuming lazy init and compilation are finished)
                if self.step == 1:
                    dist_utils.set_pg_timeouts(
                        timeout=timedelta(
                            seconds=job_config.comm.train_timeout_seconds
                        ),
                        world_mesh=self.parallel_dims.world_mesh,
                    )

        if torch.distributed.get_rank() == 0:
            logger.info("Sleeping 2 seconds for other ranks to complete")
            time.sleep(2)

        logger.info("Training completed")

    def should_continue_training(self) -> bool:
        return self.step < self.job_config.training.steps

    def state_dict(self) -> dict[str, Any]:
        return {"step": self.step, "ntokens_seen": self.ntokens_seen}

    def load_state_dict(self, state_dict: dict[str, Any]):
        self.step = state_dict["step"]
        self.ntokens_seen = state_dict["ntokens_seen"]

    def close(self) -> None:
        if self.checkpointer:
            self.checkpointer.close()
        if self.metrics_processor:
            self.metrics_processor.close()


if __name__ == "__main__":
    init_logger()
    config_manager = ConfigManager()
    config = config_manager.parse_args()
    trainer: Optional[Trainer] = None

    try:
        trainer = Trainer(config)

        if config.checkpoint.create_seed_checkpoint:
            assert (
                int(os.environ["WORLD_SIZE"]) == 1
            ), "Must create seed checkpoint using a single device, to disable sharding."
            assert (
                config.checkpoint.enable
            ), "Must enable checkpointing when creating a seed checkpoint."
            trainer.checkpointer.save(curr_step=0, last_step=True)
            logger.info("Created seed checkpoint")
        else:
            trainer.train()
    except Exception:
        if trainer:
            trainer.close()
        raise
    else:
        trainer.close()
        torch.distributed.destroy_process_group()
        logger.info("Process group destroyed")