composer.trainer.trainer#

Train models!

The trainer supports models with ComposerModel instances. The Trainer is highly customizable and can support a wide variety of workloads.

Example

Train a model and save a checkpoint:

import os
from composer import Trainer

### Create a trainer
trainer = Trainer(
    model=model,
    train_dataloader=train_dataloader,
    max_duration="1ep",
    eval_dataloader=eval_dataloader,
    optimizers=optimizer,
    schedulers=scheduler,
    device="cpu",
    validate_every_n_epochs=1,
    save_folder="checkpoints",
    save_filename="ep{epoch}.pt",
    save_interval="1ep",
    save_overwrite=True,
)

# Fit and run evaluation for 1 epoch.
# Save a checkpoint after 1 epoch as specified during trainer creation.
trainer.fit()

Load the checkpoint and resume training:

# Get the saved checkpoint filepath
checkpoint_path = trainer.saved_checkpoints.pop()[0]

# Create a new trainer with the `load_path` argument set to the checkpoint path.
trainer = Trainer(
    model=model,
    train_dataloader=train_dataloader,
    max_duration="2ep",
    eval_dataloader=eval_dataloader,
    optimizers=optimizer,
    schedulers=scheduler,
    device="cpu",
    validate_every_n_epochs=1,
    load_path=checkpoint_path,
)

# Continue training and running evaluation where the previous trainer left off
# until the new max_duration is reached.
# In this case it will be one additional epoch to reach 2 epochs total.
trainer.fit()

Classes

Trainer

Trainer for training a models with Composer algorithms.

class composer.trainer.trainer.Trainer(*, model, train_dataloader, max_duration, eval_dataloader=None, algorithms=None, optimizers=None, schedulers=None, device=None, grad_accum=1, grad_clip_norm=None, validate_every_n_batches=-1, validate_every_n_epochs=1, compute_training_metrics=False, precision=Precision.FP32, scale_schedule_ratio=1.0, step_schedulers_every_batch=None, dist_timeout=300.0, ddp_sync_strategy=None, seed=None, deterministic_mode=False, run_name=None, loggers=None, callbacks=(), progress_bar=True, log_to_console=None, console_log_level=LogLevel.EPOCH, console_stream=<_io.TextIOWrapper name='<stderr>' mode='w' encoding='utf-8'>, load_path=None, load_object_store=None, load_weights_only=False, load_strict=False, load_chunk_size=1048576, load_progress_bar=True, save_folder=None, save_filename='ep{epoch}-ba{batch}-rank{rank}', save_artifact_name='{run_name}/checkpoints/ep{epoch}-ba{batch}-rank{rank}', save_latest_filename='latest-rank{rank}', save_latest_artifact_name='{run_name}/checkpoints/latest-rank{rank}', save_overwrite=False, save_interval='1ep', save_weights_only=False, save_num_checkpoints_to_keep=-1, train_subset_num_batches=None, eval_subset_num_batches=None, deepspeed_config=False, prof_trace_handlers=None, prof_schedule=None, sys_prof_cpu=True, sys_prof_memory=False, sys_prof_disk=False, sys_prof_net=False, sys_prof_stats_thread_interval_seconds=0.5, torch_prof_folder='{run_name}/torch_traces', torch_prof_filename='rank{rank}.{batch}.pt.trace.json', torch_prof_artifact_name='{run_name}/torch_traces/rank{rank}.{batch}.pt.trace.json', torch_prof_overwrite=False, torch_prof_use_gzip=False, torch_prof_record_shapes=False, torch_prof_profile_memory=True, torch_prof_with_stack=False, torch_prof_with_flops=True, torch_prof_num_traces_to_keep=-1)[source]#

Trainer for training a models with Composer algorithms. See the Trainer guide for more information.

Parameters

model (ComposerModel) –
The model to train. Can be user-defined or one of the models included with Composer.

See also

composer.models for models built into Composer.
train_dataloader (DataLoader, DataSpec, or dict) –
The DataLoader, DataSpec, or dict of DataSpec kwargs for the training data. In order to specify custom preprocessing steps on each data batch, specify a DataSpec instead of a DataLoader.

Note

The train_dataloader should yield per-rank batches. Each per-rank batch will then be further divided based on the grad_accum parameter. For example, if the desired optimization batch size is 2048 and training is happening across 8 GPUs, then each train_dataloader should yield a batch of size 2048 / 8 = 256. If grad_accum = 2, then the per-rank batch will be divided into microbatches of size 256 / 2 = 128.
max_duration (int, str, or Time) – The maximum duration to train. Can be an integer, which will be interpreted to be epochs, a str (e.g. 1ep, or 10ba), or a Time object.
eval_dataloader (DataLoader | DataSpec | Evaluator | Sequence[Evaluator], optional) –
The DataLoader, DataSpec, Evaluator, or sequence of evaluators for the evaluation data.

To evaluate one or more specific metrics across one or more datasets, pass in an Evaluator. If a DataSpec or DataLoader is passed in, then all metrics returned by model.metrics() will be used during evaluation. None results in no evaluation. (default: None)
algorithms (Algorithm | Sequence[Algorithm], optional) –
The algorithms to use during training. If None, then no algorithms will be used. (default: None)

See also

composer.algorithms for the different algorithms built into Composer.
optimizers (Optimizer, optional) –
The optimizer. If None, will be set to DecoupledSGDW(model.parameters(), lr=0.1). (default: None)

See also

composer.optim for the different optimizers built into Composer.
schedulers (PyTorchScheduler | ComposerScheduler | Sequence[PyTorchScheduler | ComposerScheduler], optional) –
The learning rate schedulers. If [] or None, the learning rate will be constant. (default: None).

See also

composer.optim.scheduler for the different schedulers built into Composer.
device (str or Device, optional) – The device to use for training. Either cpu or gpu. (default: cpu)
grad_accum (Union[int, str], optional) –
The number of microbatches to split a per-device batch into. Gradients are summed over the microbatches per device. If set to auto, dynamically increases grad_accum if microbatch is too large for GPU. (default: 1)

Note

This is implemented by taking the batch yielded by the train_dataloader and splitting it into grad_accum sections. Each section is of size train_dataloader // grad_accum. If the batch size of the dataloader is not divisible by grad_accum, then the last section will be of size batch_size % grad_accum.
grad_clip_norm (float, optional) – The norm to clip gradient magnitudes to. Set to None for no gradient clipping. (default: None)
validate_every_n_batches (int, optional) – Compute metrics on evaluation data every N batches. Set to -1 to never validate on a batchwise frequency. (default: -1)
validate_every_n_epochs (int, optional) – Compute metrics on evaluation data every N epochs. Set to -1 to never validate on a epochwise frequency. (default: 1)
compute_training_metrics (bool, optional) – True to compute metrics on training data and False to not. (default: False)
precision (str or Precision, optional) –
Numerical precision to use for training. One of fp32, fp16 or amp (recommended). (default: Precision.FP32)

Note

fp16 only works if deepspeed_config is also provided.
scale_schedule_ratio (float, optional) –
Ratio by which to scale the training duration and learning rate schedules. E.g., 0.5 makes the schedule take half as many epochs and 2.0 makes it take twice as many epochs. 1.0 means no change. (default: 1.0)

Note

Training for less time, while rescaling the learning rate schedule, is a strong baseline approach to speeding up training. E.g., training for half duration often yields minor accuracy degradation, provided that the learning rate schedule is also rescaled to take half as long.

To see the difference, consider training for half as long using a cosine annealing learning rate schedule. If the schedule is not rescaled, training ends while the learning rate is still ~0.5 of the initial LR. If the schedule is rescaled with scale_schedule_ratio, the LR schedule would finish the entire cosine curve, ending with a learning rate near zero.
step_schedulers_every_batch (bool, optional) – By default, native PyTorch schedulers are updated every epoch, while Composer Schedulers are updated every step. Setting this to True will force schedulers to be stepped every batch, while False means schedulers stepped every epoch. None indicates the default behavior. (default: None)
dist_timeout (float, optional) – Timeout, in seconds, for initializing the distributed process group. (default: 15.0)
ddp_sync_strategy (str or DDPSyncStrategy, optional) – The strategy to use for synchronizing gradients. Leave unset to let the trainer auto-configure this. See DDPSyncStrategy for more details.
seed (int, optional) –
The seed used in randomization. If None, then a random seed will be created. (default: None)

Note

In order to get reproducible results, call the seed_all() function at the start of your script with the seed passed to the trainer. This will ensure any initialization done before the trainer init (ex. model weight initialization) also uses the provided seed.

See also

composer.utils.reproducibility for more details on reproducibility.
deterministic_mode (bool, optional) –
Run the model deterministically. (default: False)

Note

This is an experimental feature. Performance degradations expected. Certain Torch modules may not have deterministic implementations, which will result in a crash.

Note

In order to get reproducible results, call the configure_deterministic_mode() function at the start of your script. This will ensure any initialization done before the trainer init also runs deterministically.

See also

composer.utils.reproducibility for more details on reproducibility.
run_name (str, optional) –
A name for this training run. If not specified, one will be generated automatically.

See also

Logger
loggers (LoggerDestination | Sequence[LoggerDestination], optional) –
The destinations to log training information to. If None, will be set to [ProgressBarLogger()]. (default: None)

See also

composer.loggers for the different loggers built into Composer.
progress_bar (bool, optional) – Whether to show a progress bar. (default: True)
log_to_console (bool, optional) –
Whether to print logging statements to the console. (default: None)

The default behavior (when set to None) only prints logging statements when show_pbar is False.
console_log_level (LogLevel | str | (State, LogLevel) -> bool, optional) –
The maximum log level which should be printed to the console. (default: LogLevel.EPOCH)

It can either be LogLevel, a string corresponding to a LogLevel, or a callable that takes the training State and the LogLevel and returns a boolean of whether this statement should be printed.

This parameter has no effect if log_to_console is False, or is unspecified and progres_bar is True.
console_stream (TextIO | str, optional) – The stream to write to. If a string, it can either be 'stdout' or 'stderr'. (default: sys.stderr)
callbacks (Callback | Sequence[Callback], optional) –
The callbacks to run during training. If None, then no callbacks will be run. (default: None).

See also

composer.callbacks for the different callbacks built into Composer.

load_path (str, optional) –

The path format string to an existing checkpoint file.

It can be a path to a file on the local disk, a URL, or if load_object_store is set, the object name for a checkpoint in a cloud bucket.

When using Deepspeed ZeRO, checkpoints are shareded by rank. Instead of hard-coding the rank in the path, use the following format variables:

Variable	Description
`{rank}`	The global rank, as returned by `get_global_rank()`.
`{local_rank}`	The local rank of the process, as returned by `get_local_rank()`.
`{node_rank}`	The node rank, as returned by `get_node_rank()`.

For example, suppose that checkpoints are stored in the following structure:

my_model/ep1-rank0.tar
my_model/ep1-rank1.tar
my_model/ep1-rank2.tar
...

Then, load_path should be set to my_model/ep1-rank{rank}.tar, and all ranks will load the correct state.

If None then no checkpoint will be loaded. (default: None)

load_object_store (ObjectStore, optional) –

If the load_path is in an object store (i.e. AWS S3 or Google Cloud Storage), an instance of ObjectStore which will be used to retreive the checkpoint. Otherwise, if the checkpoint is a local filepath, set to None. Ignored if load_path is None. (default: None)

Example:

from composer import Trainer
from composer.utils import ObjectStore

# Create the object store provider with the specified credentials
creds = {"key": "object_store_key",
         "secret": "object_store_secret"}
store = ObjectStore(provider="s3",
                            container="my_container",
                            provider_kwargs=creds)

checkpoint_path = "./path_to_the_checkpoint_in_object_store"

# Create a trainer which will load a checkpoint from the specified object store
trainer = Trainer(
    model=model,
    train_dataloader=train_dataloader,
    max_duration="10ep",
    eval_dataloader=eval_dataloader,
    optimizers=optimizer,
    schedulers=scheduler,
    device="cpu",
    validate_every_n_epochs=1,
    load_path=checkpoint_path,
    load_object_store=store,
)

load_weights_only (bool, optional) – Whether or not to only restore the weights from the checkpoint without restoring the associated state. Ignored if load_path is None. (default: False)
load_strict (bool, optional) – Ensure that the set of weights in the checkpoint and model must exactly match. Ignored if load_path is None. (default: False)
load_chunk_size (int, optional) – Chunk size (in bytes) to use when downloading checkpoints. Ignored if load_path is either None or a local file path. (default: 1,048,675)
load_progress_bar (bool, optional) – Display the progress bar for downloading the checkpoint. Ignored if load_path is either None or a local file path. (default: True)
save_folder (str, optional) –
Format string for the folder where checkpoints are saved. If None, checkpoints will not be saved. (default: None)

See also

CheckpointSaver

Note

For fine-grained control on checkpoint saving (e.g. to save different types of checkpoints at different intervals), leave this parameter as None, and instead pass instance(s) of CheckpointSaver directly as callbacks.
save_filename (str, optional) –
A format string describing how to name checkpoints. This parameter has no effect if save_folder is None. (default: "ep{epoch}-ba{batch}-rank{rank}")

See also

CheckpointSaver
save_artifact_name (str, optional) –
A format string describing how to name checkpoints in loggers. This parameter has no effect if save_folder is None. (default: "{run_name}/checkpoints/ep{epoch}-ba{batch}-rank{rank}")

See also

CheckpointSaver
save_latest_filename (str, optional) –
A format string for the name of a symlink (relative to save_folder) that points to the last saved checkpoint. This parameter has no effect if save_folder is None. To disable symlinking, set this to None. (default: "latest-rank{rank}")

See also

CheckpointSaver
save_latest_artifact_name (str, optional) –
A format string describing how to name symlinks in loggers. This parameter has no effect if save_folder, save_latest_filename, or save_artifact_name are None. To disable symlinking in logger, set this or save_latest_filename to None. (default: "{run_name}/checkpoints/latest-rank{rank}")

See also

CheckpointSaver
save_overwrite (bool, optional) –
Whether existing checkpoints should be overridden. This parameter has no effect if save_folder is None. (default: False)

See also

CheckpointSaver
save_interval (Time | str | int | (State, Event) -> bool) –
A Time, time-string, integer (in epochs), or a function that takes (state, event) and returns a boolean whether a checkpoint should be saved. This parameter has no effect if save_folder is None. (default: '1ep')

See also

CheckpointSaver
save_weights_only (bool, optional) –
Whether to save only the model weights instead of the entire training state. This parameter has no effect if save_folder is None. (default: False)

See also

CheckpointSaver
save_num_checkpoints_to_keep (int, optional) –
The number of checkpoints to keep locally. The oldest checkpoints are removed first. Set to -1 to keep all checkpoints locally. (default: -1)

Checkpoints will be removed after they have been logged as a file artifact. For example, when this callback is used in conjunction with the ObjectStoreLogger, set this parameter to 0 to immediately delete checkpoints from the local disk after they have been uploaded to the object store.

This parameter only controls how many checkpoints are kept locally; checkpoints are not deleted from artifact stores.
train_subset_num_batches (int, optional) – If specified, finish every epoch early after training on this many batches. This parameter has no effect if it is greater than len(train_dataloader). If None, then the entire dataloader will be iterated over. (default: None)
eval_subset_num_batches (int, optional) – If specified, evaluate on this many batches. This parameter has no effect if it is greater than len(eval_dataloader). If None, then the entire dataloader will be iterated over. (default: None)
deepspeed_config (bool or Dict[str, Any], optional) – Configuration for DeepSpeed, formatted as a JSON according to DeepSpeed’s documentation. If True is provided, the trainer will initialize the DeepSpeed engine with an empty config {}. If False is provided, deepspeed will not be used. (default: False)

prof_schedule ((State) -> ProfilerAction, optional) –

The profiler scheduler.

Must be specified in conjunction with prof_trace_handlers to use the profiler.

from composer.trainer import Trainer
from composer.profiler import JSONTraceHandler, cyclic_schedule

trainer = Trainer(
    ...,
    prof_trace_handlers=JSONTraceHandler(
        folder='traces',
    ),
    prof_schedule=cyclic_schedule(
        skip_first=1,
        wait=0,
        warmup=1,
        active=4,
        repeat=1,
    ),
)