mirror of
https://github.com/huggingface/accelerate.git
synced 2025-10-20 18:13:46 +08:00
726140cad2
* v1 * More testing, need to try on H100 * Bigger batch for h100 test * test tweak * Fixup all tests! * Bookmark * Fix issues, working now * rm num samples * Uncomment * Give stateful dl end of dl * Make skip DL stateful * Migrate to update_state_dict * try/finally * Add comments to test * rm comment * Document * refactor out for eventual override * Doc nit * Brute force it
325 lines
13 KiB
Python
325 lines
13 KiB
Python
# Copyright 2021 The HuggingFace Inc. team. All rights reserved.
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
import argparse
|
|
import os
|
|
|
|
import evaluate
|
|
import torch
|
|
from datasets import load_dataset
|
|
from torch.optim import AdamW
|
|
from torch.utils.data import DataLoader
|
|
from transformers import AutoModelForSequenceClassification, AutoTokenizer, get_linear_schedule_with_warmup, set_seed
|
|
|
|
from accelerate import Accelerator, DataLoaderConfiguration, DistributedType
|
|
|
|
|
|
########################################################################
|
|
# This is a fully working simple example to use Accelerate
|
|
#
|
|
# This example trains a Bert base model on GLUE MRPC
|
|
# in any of the following settings (with the same script):
|
|
# - single CPU or single GPU
|
|
# - multi GPUS (using PyTorch distributed mode)
|
|
# - (multi) TPUs
|
|
# - fp16 (mixed-precision) or fp32 (normal precision)
|
|
#
|
|
# This example also demonstrates the checkpointing and sharding capabilities
|
|
#
|
|
# To run it in each of these various modes, follow the instructions
|
|
# in the readme for examples:
|
|
# https://github.com/huggingface/accelerate/tree/main/examples
|
|
#
|
|
########################################################################
|
|
|
|
|
|
MAX_GPU_BATCH_SIZE = 16
|
|
EVAL_BATCH_SIZE = 32
|
|
|
|
|
|
def training_function(config, args):
|
|
# Initialize accelerator
|
|
dataloader_config = DataLoaderConfiguration(use_stateful_dataloader=args.use_stateful_dataloader)
|
|
if args.with_tracking:
|
|
accelerator = Accelerator(
|
|
cpu=args.cpu,
|
|
mixed_precision=args.mixed_precision,
|
|
dataloader_config=dataloader_config,
|
|
log_with="all",
|
|
project_dir=args.project_dir,
|
|
)
|
|
else:
|
|
accelerator = Accelerator(
|
|
cpu=args.cpu, mixed_precision=args.mixed_precision, dataloader_config=dataloader_config
|
|
)
|
|
|
|
if hasattr(args.checkpointing_steps, "isdigit"):
|
|
if args.checkpointing_steps == "epoch":
|
|
checkpointing_steps = args.checkpointing_steps
|
|
elif args.checkpointing_steps.isdigit():
|
|
checkpointing_steps = int(args.checkpointing_steps)
|
|
else:
|
|
raise ValueError(
|
|
f"Argument `checkpointing_steps` must be either a number or `epoch`. `{args.checkpointing_steps}` passed."
|
|
)
|
|
else:
|
|
checkpointing_steps = None
|
|
# Sample hyper-parameters for learning rate, batch size, seed and a few other HPs
|
|
lr = config["lr"]
|
|
num_epochs = int(config["num_epochs"])
|
|
seed = int(config["seed"])
|
|
batch_size = int(config["batch_size"])
|
|
|
|
# We need to initialize the trackers we use, and also store our configuration
|
|
if args.with_tracking:
|
|
run = os.path.split(__file__)[-1].split(".")[0]
|
|
accelerator.init_trackers(run, config)
|
|
|
|
tokenizer = AutoTokenizer.from_pretrained("bert-base-cased")
|
|
datasets = load_dataset("glue", "mrpc")
|
|
metric = evaluate.load("glue", "mrpc")
|
|
|
|
def tokenize_function(examples):
|
|
# max_length=None => use the model max length (it's actually the default)
|
|
outputs = tokenizer(examples["sentence1"], examples["sentence2"], truncation=True, max_length=None)
|
|
return outputs
|
|
|
|
# Apply the method we just defined to all the examples in all the splits of the dataset
|
|
# starting with the main process first:
|
|
with accelerator.main_process_first():
|
|
tokenized_datasets = datasets.map(
|
|
tokenize_function,
|
|
batched=True,
|
|
remove_columns=["idx", "sentence1", "sentence2"],
|
|
)
|
|
|
|
# We also rename the 'label' column to 'labels' which is the expected name for labels by the models of the
|
|
# transformers library
|
|
tokenized_datasets = tokenized_datasets.rename_column("label", "labels")
|
|
|
|
# If the batch size is too big we use gradient accumulation
|
|
gradient_accumulation_steps = 1
|
|
if batch_size > MAX_GPU_BATCH_SIZE and accelerator.distributed_type != DistributedType.XLA:
|
|
gradient_accumulation_steps = batch_size // MAX_GPU_BATCH_SIZE
|
|
batch_size = MAX_GPU_BATCH_SIZE
|
|
|
|
def collate_fn(examples):
|
|
# On TPU it's best to pad everything to the same length or training will be very slow.
|
|
max_length = 128 if accelerator.distributed_type == DistributedType.XLA else None
|
|
# When using mixed precision we want round multiples of 8/16
|
|
if accelerator.mixed_precision == "fp8":
|
|
pad_to_multiple_of = 16
|
|
elif accelerator.mixed_precision != "no":
|
|
pad_to_multiple_of = 8
|
|
else:
|
|
pad_to_multiple_of = None
|
|
|
|
return tokenizer.pad(
|
|
examples,
|
|
padding="longest",
|
|
max_length=max_length,
|
|
pad_to_multiple_of=pad_to_multiple_of,
|
|
return_tensors="pt",
|
|
)
|
|
|
|
# Instantiate dataloaders.
|
|
train_dataloader = DataLoader(
|
|
tokenized_datasets["train"], shuffle=True, collate_fn=collate_fn, batch_size=batch_size
|
|
)
|
|
eval_dataloader = DataLoader(
|
|
tokenized_datasets["validation"], shuffle=False, collate_fn=collate_fn, batch_size=EVAL_BATCH_SIZE
|
|
)
|
|
|
|
set_seed(seed)
|
|
|
|
# Instantiate the model (we build the model here so that the seed also control new weights initialization)
|
|
model = AutoModelForSequenceClassification.from_pretrained("bert-base-cased", return_dict=True)
|
|
|
|
# We could avoid this line since the accelerator is set with `device_placement=True` (default value).
|
|
# Note that if you are placing tensors on devices manually, this line absolutely needs to be before the optimizer
|
|
# creation otherwise training will not work on TPU (`accelerate` will kindly throw an error to make us aware of that).
|
|
model = model.to(accelerator.device)
|
|
|
|
# Instantiate optimizer
|
|
optimizer = AdamW(params=model.parameters(), lr=lr)
|
|
|
|
# Instantiate scheduler
|
|
lr_scheduler = get_linear_schedule_with_warmup(
|
|
optimizer=optimizer,
|
|
num_warmup_steps=100,
|
|
num_training_steps=(len(train_dataloader) * num_epochs) // gradient_accumulation_steps,
|
|
)
|
|
|
|
# Prepare everything
|
|
# There is no specific order to remember, we just need to unpack the objects in the same order we gave them to the
|
|
# prepare method.
|
|
model, optimizer, train_dataloader, eval_dataloader, lr_scheduler = accelerator.prepare(
|
|
model, optimizer, train_dataloader, eval_dataloader, lr_scheduler
|
|
)
|
|
|
|
# We need to keep track of how many total steps we have iterated over
|
|
overall_step = 0
|
|
# We also need to keep track of the stating epoch so files are named properly
|
|
starting_epoch = 0
|
|
|
|
# Potentially load in the weights and states from a previous save
|
|
if args.resume_from_checkpoint:
|
|
if args.resume_from_checkpoint is not None or args.resume_from_checkpoint != "":
|
|
accelerator.print(f"Resumed from checkpoint: {args.resume_from_checkpoint}")
|
|
accelerator.load_state(args.resume_from_checkpoint)
|
|
path = os.path.basename(args.resume_from_checkpoint)
|
|
else:
|
|
# Get the most recent checkpoint
|
|
dirs = [f.name for f in os.scandir(os.getcwd()) if f.is_dir()]
|
|
dirs.sort(key=os.path.getctime)
|
|
path = dirs[-1] # Sorts folders by date modified, most recent checkpoint is the last
|
|
# Extract `epoch_{i}` or `step_{i}`
|
|
training_difference = os.path.splitext(path)[0]
|
|
|
|
if "epoch" in training_difference:
|
|
starting_epoch = int(training_difference.replace("epoch_", "")) + 1
|
|
resume_step = None
|
|
else:
|
|
resume_step = int(training_difference.replace("step_", ""))
|
|
starting_epoch = resume_step // len(train_dataloader)
|
|
resume_step -= starting_epoch * len(train_dataloader)
|
|
|
|
# Now we train the model
|
|
for epoch in range(starting_epoch, num_epochs):
|
|
model.train()
|
|
if args.with_tracking:
|
|
total_loss = 0
|
|
if args.resume_from_checkpoint and epoch == starting_epoch and resume_step is not None:
|
|
# We need to skip steps until we reach the resumed step
|
|
if not args.use_stateful_dataloader:
|
|
active_dataloader = accelerator.skip_first_batches(train_dataloader, resume_step)
|
|
else:
|
|
active_dataloader = train_dataloader
|
|
overall_step += resume_step
|
|
else:
|
|
# After the first iteration though, we need to go back to the original dataloader
|
|
active_dataloader = train_dataloader
|
|
for step, batch in enumerate(active_dataloader):
|
|
# We could avoid this line since we set the accelerator with `device_placement=True`.
|
|
batch.to(accelerator.device)
|
|
outputs = model(**batch)
|
|
loss = outputs.loss
|
|
loss = loss / gradient_accumulation_steps
|
|
# We keep track of the loss at each epoch
|
|
if args.with_tracking:
|
|
total_loss += loss.detach().float()
|
|
accelerator.backward(loss)
|
|
if step % gradient_accumulation_steps == 0:
|
|
optimizer.step()
|
|
lr_scheduler.step()
|
|
optimizer.zero_grad()
|
|
|
|
overall_step += 1
|
|
|
|
if isinstance(checkpointing_steps, int):
|
|
output_dir = f"step_{overall_step}"
|
|
if overall_step % checkpointing_steps == 0:
|
|
if args.output_dir is not None:
|
|
output_dir = os.path.join(args.output_dir, output_dir)
|
|
accelerator.save_state(output_dir)
|
|
|
|
model.eval()
|
|
for step, batch in enumerate(eval_dataloader):
|
|
# We could avoid this line since we set the accelerator with `device_placement=True`.
|
|
batch.to(accelerator.device)
|
|
with torch.no_grad():
|
|
outputs = model(**batch)
|
|
predictions = outputs.logits.argmax(dim=-1)
|
|
predictions, references = accelerator.gather_for_metrics((predictions, batch["labels"]))
|
|
metric.add_batch(
|
|
predictions=predictions,
|
|
references=references,
|
|
)
|
|
|
|
eval_metric = metric.compute()
|
|
# Use accelerator.print to print only on the main process.
|
|
accelerator.print(f"epoch {epoch}:", eval_metric)
|
|
if args.with_tracking:
|
|
accelerator.log(
|
|
{
|
|
"accuracy": eval_metric["accuracy"],
|
|
"f1": eval_metric["f1"],
|
|
"train_loss": total_loss.item() / len(train_dataloader),
|
|
"epoch": epoch,
|
|
},
|
|
step=epoch,
|
|
)
|
|
|
|
if checkpointing_steps == "epoch":
|
|
output_dir = f"epoch_{epoch}"
|
|
if args.output_dir is not None:
|
|
output_dir = os.path.join(args.output_dir, output_dir)
|
|
accelerator.save_state(output_dir)
|
|
|
|
accelerator.end_training()
|
|
|
|
|
|
def main():
|
|
parser = argparse.ArgumentParser(description="Simple example of training script.")
|
|
parser.add_argument(
|
|
"--mixed_precision",
|
|
type=str,
|
|
default=None,
|
|
choices=["no", "fp16", "bf16", "fp8"],
|
|
help="Whether to use mixed precision. Choose"
|
|
"between fp16 and bf16 (bfloat16). Bf16 requires PyTorch >= 1.10."
|
|
"and an Nvidia Ampere GPU.",
|
|
)
|
|
parser.add_argument("--cpu", action="store_true", help="If passed, will train on the CPU.")
|
|
parser.add_argument(
|
|
"--checkpointing_steps",
|
|
type=str,
|
|
default=None,
|
|
help="Whether the various states should be saved at the end of every n steps, or 'epoch' for each epoch.",
|
|
)
|
|
parser.add_argument(
|
|
"--resume_from_checkpoint",
|
|
type=str,
|
|
default=None,
|
|
help="If the training should continue from a checkpoint folder.",
|
|
)
|
|
parser.add_argument(
|
|
"--use_stateful_dataloader",
|
|
action="store_true",
|
|
help="If the dataloader should be a resumable stateful dataloader.",
|
|
)
|
|
parser.add_argument(
|
|
"--with_tracking",
|
|
action="store_true",
|
|
help="Whether to load in all available experiment trackers from the environment and use them for logging.",
|
|
)
|
|
parser.add_argument(
|
|
"--output_dir",
|
|
type=str,
|
|
default=".",
|
|
help="Optional save directory where all checkpoint folders will be stored. Default is the current working directory.",
|
|
)
|
|
parser.add_argument(
|
|
"--project_dir",
|
|
type=str,
|
|
default="logs",
|
|
help="Location on where to store experiment tracking logs` and relevent project information",
|
|
)
|
|
args = parser.parse_args()
|
|
config = {"lr": 2e-5, "num_epochs": 3, "seed": 42, "batch_size": 16}
|
|
training_function(config, args)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
main()
|