def main():
"""
Main function to run the training and evaluation of a causal language model.
This function sets up the training environment, loads the datasets, tokenizes the data,
initializes the model and tokenizer, and runs the training and evaluation loops.
It also handles distributed training setup, logging, and configuration loading.
"""
# hfloglevel = "INFO" if rank == 0 else "ERROR"
# logging.getLogger("datasets").setLevel(hfloglevel)
rank = ezpz.setup_torch()
try:
import wandb
except (ImportError, ModuleNotFoundError):
wandb = None # type: ignore
try:
import evaluate
except (ImportError, ModuleNotFoundError):
evaluate = None # type: ignore
print(
'"evaluate" library is not installed. '
"We will continue without running evaluations. "
'Please install it using "pip install evaluate" to run evaluations'
)
args = parse_args()
data_args = args["data"]
model_args = args["model"]
training_args = args["training"]
dsconfig_fp = Path(training_args.deepspeed) if training_args.deepspeed else None
ds_config = ezpz.configs.load_ds_config(dsconfig_fp)
if training_args.optim is not None and "optimizer" in ds_config:
logger.warning(
f"Overriding optimizer in deepspeed config with {training_args.optim}"
)
_ = ds_config.pop("optimizer")
# Detecting last checkpoint.
last_checkpoint = None
if (
os.path.isdir(training_args.output_dir)
and training_args.do_train
and not training_args.overwrite_output_dir
):
last_checkpoint = get_last_checkpoint(training_args.output_dir)
if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
raise ValueError(
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
"Use --overwrite_output_dir to overcome."
)
elif (
last_checkpoint is not None and training_args.resume_from_checkpoint is None
):
logger.info(
f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
)
# Set seed before initializing model.
set_seed(training_args.seed)
# from datasets import datasets.load_dataset
# Get the datasets: you can either provide your own CSV/JSON/TXT training and evaluation files (see below)
# or just provide the name of one of the public datasets available on the hub at https://huggingface.co/datasets/
# (the dataset will be downloaded automatically from the datasets Hub).
#
# For CSV/JSON files, this script will use the column called 'text' or the first column if no column called
# 'text' is found. You can easily tweak this behavior (see below).
#
# In distributed training, the datasets.load_dataset function guarantee that only one local process can concurrently
# download the dataset.
train_split_name = data_args.train_split_name
validation_split_name = data_args.validation_split_name
test_split_name = data_args.test_split_name
if data_args.dataset_name is not None:
# Downloading and loading a dataset from the hub.
# dataset = datasets.load_dataset(
# data_args.dataset_name,
# data_args.dataset_config_name,
# cache_dir=model_args.cache_dir,
# token=model_args.token,
# streaming=data_args.streaming,
# trust_remote_code=model_args.trust_remote_code,
# )
raw_datasets = split_dataset(
model_args,
data_args,
train_split_name=train_split_name,
validation_split_name=validation_split_name,
)
else:
data_files = {}
dataset_args = {}
if data_args.train_file is not None:
data_files[train_split_name] = data_args.train_file
if data_args.validation_file is not None:
data_files[validation_split_name] = data_args.validation_file
extension = (
data_args.train_file.split(".")[-1]
if data_args.train_file is not None
else data_args.validation_file.split(".")[-1]
)
if extension == "txt":
extension = "text"
dataset_args["keep_linebreaks"] = data_args.keep_linebreaks
raw_datasets = datasets.load_dataset(
extension,
data_files=data_files,
cache_dir=model_args.cache_dir,
token=model_args.token,
**dataset_args,
)
# If no validation data is there, validation_split_percentage will be used to divide the dataset.
if validation_split_name not in raw_datasets.keys(): # type:ignore
raw_datasets[validation_split_name] = datasets.load_dataset( # type:ignore
extension,
data_files=data_files,
split=f"{train_split_name}[:{data_args.validation_split_percentage}%]",
cache_dir=model_args.cache_dir,
token=model_args.token,
**dataset_args,
)
raw_datasets[train_split_name] = datasets.load_dataset( # type:ignore
extension,
data_files=data_files,
split=f"{train_split_name}[{data_args.validation_split_percentage}%:]",
cache_dir=model_args.cache_dir,
token=model_args.token,
**dataset_args,
)
# See more about loading any type of standard or custom dataset (from
# files, python dict, pandas DataFrame, etc) at:
# https://huggingface.co/docs/datasets/loading_datasets.
config_kwargs = {
"cache_dir": model_args.cache_dir,
"revision": model_args.model_revision,
"token": model_args.token,
"trust_remote_code": model_args.trust_remote_code,
}
if model_args.config_name:
config = AutoConfig.from_pretrained(model_args.config_name, **config_kwargs)
elif model_args.model_name_or_path:
config = AutoConfig.from_pretrained(
model_args.model_name_or_path, **config_kwargs
)
else:
config = CONFIG_MAPPING[model_args.model_type]()
logger.warning("You are instantiating a new config instance from scratch.")
if model_args.config_overrides is not None:
logger.info(f"Overriding config: {model_args.config_overrides}")
config.update_from_string(model_args.config_overrides)
logger.info(f"New config: {config}")
tokenizer_kwargs = {
"cache_dir": model_args.cache_dir,
"use_fast": model_args.use_fast_tokenizer,
"revision": model_args.model_revision,
"token": model_args.token,
"trust_remote_code": model_args.trust_remote_code,
}
if model_args.tokenizer_name:
tokenizer = AutoTokenizer.from_pretrained(
model_args.tokenizer_name, **tokenizer_kwargs
)
elif model_args.model_name_or_path:
tokenizer = AutoTokenizer.from_pretrained(
model_args.model_name_or_path, **tokenizer_kwargs
)
else:
raise ValueError(
"You are instantiating a new tokenizer from scratch. This is not supported by this script. "
"You can do it from another script, save it, and load it from here, using --tokenizer_name."
)
# Load pretrained model and tokenizer
#
# Distributed training:
# The .from_pretrained methods guarantee that only one local process can concurrently
# download model & vocab.
if model_args.model_name_or_path:
torch_dtype = (
model_args.torch_dtype
if model_args.torch_dtype in ["auto", None]
else getattr(torch, model_args.torch_dtype)
)
# mm = AutoModelForCausalLM.
model = AutoModelForCausalLM.from_pretrained( # type:ignore
model_args.model_name_or_path,
from_tf=bool(".ckpt" in model_args.model_name_or_path),
config=config,
cache_dir=model_args.cache_dir,
revision=model_args.model_revision,
token=model_args.token,
trust_remote_code=model_args.trust_remote_code,
torch_dtype=torch_dtype,
low_cpu_mem_usage=model_args.low_cpu_mem_usage,
)
else:
model = AutoModelForCausalLM.from_config( # type:ignore
config, trust_remote_code=model_args.trust_remote_code
)
n_params = sum({p.data_ptr(): p.numel() for p in model.parameters()}.values())
logger.info(
f"Training new model from scratch - Total size={n_params / 2**20:.2f}M params"
)
if wandb is not None and getattr(wandb, "run", None) is not None:
wandb.watch(model, log="all")
# We resize the embeddings only when necessary to avoid index errors. If you are creating a model from scratch
# on a small vocab and want a smaller embedding size, remove this test.
embedding_size = model.get_input_embeddings().weight.shape[0]
if len(tokenizer) > embedding_size:
model.resize_token_embeddings(len(tokenizer))
# Preprocessing the datasets.
# First we tokenize all the texts.
if training_args.do_train:
column_names = list(raw_datasets[train_split_name].features) # type:ignore
else:
column_names = list(raw_datasets[validation_split_name].features) # type:ignore
text_column_name = "text" if "text" in column_names else column_names[0]
# since this will be pickled to avoid _LazyModule error in Hasher force logger loading before tokenize_function
# tok_logger = transformers.utils.logging.get_logger(
# "transformers.tokenization_utils_base"
# )
def tokenize_function(examples):
# with CaptureLogger(tok_logger) as cl:
output = tokenizer(examples[text_column_name])
# clm input could be much much longer than block_size
# if "Token indices sequence length is longer than the" in cl.out:
# logger.warning(
# "^^^^^^^^^^^^^^^^ Please ignore the warning above - this long input will be chunked into smaller bits"
# " before being passed to the model."
# )
return output
with training_args.main_process_first(desc="dataset map tokenization"):
if not data_args.streaming:
tokenized_datasets = raw_datasets.map(
tokenize_function,
batched=True,
num_proc=data_args.preprocessing_num_workers, # type:ignore
remove_columns=column_names,
load_from_cache_file=not data_args.overwrite_cache, # type:ignore
desc="Running tokenizer on dataset", # type:ignore
)
else:
tokenized_datasets = raw_datasets.map(
tokenize_function,
batched=True,
remove_columns=column_names,
)
if hasattr(config, "max_position_embeddings"):
max_pos_embeddings = config.max_position_embeddings
else:
# Define a default value if the attribute is missing in the config.
max_pos_embeddings = 1024
logger.warning(
f"Config {config} does not have 'max_position_embeddings' attribute. "
f"Using default value of {max_pos_embeddings}."
)
if data_args.block_size is None:
block_size = tokenizer.model_max_length
if block_size > max_pos_embeddings:
logger.warning(
f"The tokenizer picked seems to have a very large `model_max_length` ({tokenizer.model_max_length}). "
f"Using block_size={min(1024, max_pos_embeddings)} instead. You can change that default value by passing --block_size xxx."
)
if max_pos_embeddings > 0:
block_size = min(1024, max_pos_embeddings)
else:
block_size = 1024
else:
if data_args.block_size > tokenizer.model_max_length:
logger.warning(
f"The block_size passed ({data_args.block_size}) is larger than the maximum length for the model "
f"({tokenizer.model_max_length}). Using block_size={tokenizer.model_max_length}."
)
block_size = min(data_args.block_size, tokenizer.model_max_length)
# Main data processing function that will concatenate all texts from our dataset and generate chunks of block_size.
def group_texts(examples):
# Concatenate all texts.
concatenated_examples = {k: list(chain(*examples[k])) for k in examples.keys()}
total_length = len(concatenated_examples[list(examples.keys())[0]])
# We drop the small remainder, and if the total_length < block_size we
# exclude this batch and return an empty dict.
# We could add padding if the model supported it instead of this drop,
# you can customize this part to your needs.
total_length = (total_length // block_size) * block_size
# Split by chunks of max_len.
result = {
k: [t[i : i + block_size] for i in range(0, total_length, block_size)]
for k, t in concatenated_examples.items()
}
result["labels"] = result["input_ids"].copy()
return result
# Note that with `batched=True`, this map processes 1,000 texts together,
# so group_texts throws away a remainder for each of those groups of 1,000
# texts. You can adjust that batch_size here but a higher value might be
# slower to preprocess.
#
# To speed up this part, we use multiprocessing. See the documentation of
# the map method for more information:
# https://huggingface.co/docs/datasets/process#map
with training_args.main_process_first(desc="grouping texts together"):
if not data_args.streaming:
lm_datasets = tokenized_datasets.map(
group_texts,
batched=True,
num_proc=data_args.preprocessing_num_workers, # type:ignore
load_from_cache_file=not data_args.overwrite_cache, # type:ignore
desc=f"Grouping texts in chunks of {block_size}", # type:ignore
)
else:
lm_datasets = tokenized_datasets.map(
group_texts,
batched=True,
)
train_dataset = None
if training_args.do_train:
if "train" not in tokenized_datasets:
raise ValueError("--do_train requires a train dataset")
train_dataset = lm_datasets["train"] # type:ignore
assert isinstance(train_dataset, (datasets.Dataset, datasets.IterableDataset))
if data_args.max_train_samples is not None:
if isinstance(train_dataset, datasets.IterableDataset):
train_dataset = train_dataset.take(data_args.max_train_samples)
elif isinstance(train_dataset, datasets.Dataset):
max_train_samples = min(
train_dataset.num_rows, data_args.max_train_samples
)
train_dataset = train_dataset.select(range(max_train_samples))
# available_splits = train_dataset.info.splits
# if (train_split := getattr(available_splits, "train")) is not None:
# if data_args.max_train_samples is not None:
# max_train_samples = min(
# train_split.num_examples,
# data_args.max_train_samples,
# )
# train_dataset = train_dataset.select( # type:ignore
# range(max_train_samples)
# )
# if available_splits is not None and "train" not in available_splits:
# if isinstance(train_dataset, datasets.IterableDataset):
# train_dataset = train_dataset.take(max_eval_samples)
# else:
# eval_dataset = eval_dataset.select(range(max_eval_samples))
# if training_args.eval_on_start and evaluate is not None:
# metric = evaluate.load("accuracy", cache_dir=model_args.cache_dir)
#
# def compute_metrics(eval_preds):
# preds, labels = eval_preds
# # preds have the same shape as the labels, after the argmax(-1) has been calculated
# # by preprocess_logits_for_metrics but we need to shift the labels
# labels = labels[:, 1:].reshape(-1)
# preds = preds[:, :-1].reshape(-1)
# predictions = decode_predictions(tokenizer, preds)
# predictions_df = pd.DataFrame(predictions)
# if wandb is not None and getattr(wandb, "run", None) is not None:
# records_table = wandb.Table(dataframe=predictions_df)
# # log the table to wandb
# assert wandb is not None and wandb.run is not None
# wandb.log({"sample_predictions": records_table})
# return metric.compute(predictions=preds, references=labels)
eval_dataset = None
if training_args.do_eval:
assert evaluate is not None
if validation_split_name not in tokenized_datasets:
raise ValueError("--do_eval requires a validation dataset")
eval_dataset = lm_datasets[validation_split_name] # type: ignore
if data_args.max_eval_samples is not None:
# max_eval_samples = min(
# # len(eval_dataset),
# # eval_dataset.info.splits["train"].num_examples, # type:ignore
# data_args.max_eval_samples,
# )
if isinstance(eval_dataset, datasets.IterableDataset):
eval_dataset = eval_dataset.take(data_args.max_eval_samples)
else:
eval_dataset = eval_dataset.select(
range(data_args.max_eval_samples)
) # type:ignore
def preprocess_logits_for_metrics(logits, labels):
if isinstance(logits, tuple):
# Depending on the model and config, logits may contain extra tensors,
# like past_key_values, but logits always come first
logits = logits[0]
return logits.argmax(dim=-1)
# if evaluate is not None:
metric = evaluate.load("accuracy", cache_dir=model_args.cache_dir)
def compute_metrics(
eval_preds: tuple[torch.Tensor, torch.Tensor],
) -> dict | None:
preds, labels = eval_preds
# preds have the same shape as the labels, after the argmax(-1) has been calculated
# by preprocess_logits_for_metrics but we need to shift the labels
labels = labels[:, 1:].reshape(-1)
preds = preds[:, :-1].reshape(-1)
# predictions = decode_predictions(tokenizer, preds)
# predictions_df = pd.DataFrame(predictions)
# if wandb is not None and getattr(wandb, "run", None) is not None:
# records_table = wandb.Table(dataframe=predictions_df)
# # log the table to wandb
# assert wandb is not None and wandb.run is not None
# wandb.log({"sample_predictions": records_table})
# return metric.compute(predictions=preds, references=labels)
return metric.compute(predictions=preds, references=labels)
if rank == 0 and wandb is not None:
wandb.run.watch(model, log="all") # type:ignore
# Initialize our Trainer
trainer = Trainer(
model=model,
args=training_args,
train_dataset=train_dataset if training_args.do_train else None, # type:ignore
eval_dataset=eval_dataset if training_args.do_eval else None, # type:ignore
processing_class=tokenizer,
# Data collator will default to DataCollatorWithPadding, so we change it.
data_collator=default_data_collator,
compute_metrics=(
compute_metrics # type:ignore
if training_args.do_eval and not is_torch_xla_available() # type:ignore
else None
),
preprocess_logits_for_metrics=(
preprocess_logits_for_metrics # type:ignore
if training_args.do_eval and not is_torch_xla_available()
else None
),
)
# if wandb is not None and getattr(wandb, "run", None) is not None:
# # from transformers.integrations.integration_utils import W
# # from wandb.integration import WandbEvalCallback
#
# # evals_callback = WandbEvalCallback(trainer, tokenizer)
# from ezpz.integrations import WandbPredictionProgressCallback
#
# callback = WandbPredictionProgressCallback(
# trainer=trainer,
# tokenizer=tokenizer,
# val_dataset=(eval_dataset if training_args.do_eval else None),
# num_samples=10,
# freq=2,
# )
# trainer.add_callback(callback)
assert any(
[training_args.do_train, training_args.do_eval]
), "Nothing to do! Set --do_train or --do_eval."
# Training
if training_args.do_train:
assert isinstance(trainer, Trainer)
checkpoint = None
if training_args.resume_from_checkpoint is not None:
checkpoint = training_args.resume_from_checkpoint
elif last_checkpoint is not None:
checkpoint = last_checkpoint
train_result = trainer.train(resume_from_checkpoint=checkpoint)
trainer.save_model() # Saves the tokenizer too for easy upload
trainer.save_state()
metrics = train_result.metrics
max_train_samples = (
data_args.max_train_samples
if data_args.max_train_samples is not None
else train_dataset.info.splits["train"].num_examples # type:ignore
)
assert train_dataset is not None
metrics["train_samples"] = min(
max_train_samples,
train_dataset.info.splits["train"].num_examples, # type:ignore
) # type:ignore
trainer.log_metrics("train", metrics)
trainer.save_metrics("train", metrics)
# Evaluation
if training_args.do_eval:
assert isinstance(trainer, Trainer) and callable(
getattr(trainer, "evaluate")
), (
"Trainer must be an instance of `transformers.Trainer` "
"and have an `evaluate` method."
)
assert trainer.evaluate is not None and callable(
trainer.evaluate
), "Trainer must have an `evaluate` method."
metrics = trainer.evaluate()
max_eval_samples = (
data_args.max_eval_samples
if data_args.max_eval_samples is not None
else eval_dataset.info.splits["train"].num_examples # type:ignore
)
metrics["eval_samples"] = min(
max_eval_samples,
eval_dataset.info.splits["train"].num_examples, # type:ignore
) # type:ignore
try:
perplexity = math.exp(metrics["eval_loss"])
except OverflowError:
perplexity = float("inf")
metrics["perplexity"] = perplexity
trainer.log_metrics("eval", metrics)
trainer.save_metrics("eval", metrics)
if wandb is not None and wandb.run is not None:
wandb.log({f"eval/{k}": v for k, v in metrics.items()})
# model.eval()
# losses = []
# assert eval_dataset is not None
# for step, batch in enumerate(eval_dataloader):
# with torch.no_grad():
# outputs = model(**batch)
# loss = outputs.loss
# losses.append(loss.item())
# # Evaluation
# if training_args.do_eval:
# import datasets
# import pandas as pd
#
# logger.info("*** Evaluate ***")
# assert eval_dataset is not None, (
# "eval_dataset must be defined for evaluation."
# )
#
# metrics = trainer.evaluate()
#
# max_eval_samples = (
# data_args.max_eval_samples
# if data_args.max_eval_samples is not None
# else len(eval_dataset) # type:ignore
# )
# metrics["eval_samples"] = min(max_eval_samples, len(eval_dataset)) # type:ignore
# try:
# perplexity = math.exp(metrics["eval_loss"])
# except OverflowError:
# perplexity = float("inf")
# metrics["perplexity"] = perplexity
# trainer.log_metrics("eval", metrics)
# trainer.save_metrics("eval", metrics)
#
# # sample_dataset = eval_dataset.take(10)
# # assert sample_dataset is not None
# # # generate predictions
# # if sample_dataset is not None and isinstance(sample_dataset, datasets.Dataset):
# # predictions = trainer.predict(sample_dataset)
# # # decode predictions and labels
# # predictions = decode_predictions(tokenizer, predictions)
# # # add predictions to a wandb.Table
# # predictions_df = pd.DataFrame(predictions)
# # # predictions_df["epoch"] = state.epoch
# # if wandb is not None and getattr(wandb, "run", None) is not None:
# # records_table = wandb.Table(dataframe=predictions_df)
# # # log the table to wandb
# # assert wandb is not None and wandb.run is not None
# # wandb.log({"sample_predictions": records_table})
kwargs = {
"finetuned_from": model_args.model_name_or_path,
"tasks": "text-generation",
}
if data_args.dataset_name is not None:
kwargs["dataset_tags"] = data_args.dataset_name
if data_args.dataset_config_name is not None:
kwargs["dataset_args"] = data_args.dataset_config_name
kwargs["dataset"] = (
f"{data_args.dataset_name} {data_args.dataset_config_name}"
)
else:
kwargs["dataset"] = data_args.dataset_name
if wandb is not None and getattr(wandb, "run", None) is not None:
wandb.config.update(kwargs)
if training_args.push_to_hub:
trainer.push_to_hub(**kwargs)
else:
trainer.create_model_card(**kwargs)