amyeroberts
commented
3 months ago cc @muellerzr @pacman100
Open yuane4 opened 3 months ago
amyeroberts
commented
3 months ago cc @muellerzr @pacman100
muellerzr
commented
3 months ago @yuane4 can you please share your entire script?
yuane4
commented
3 months ago yes of course, is there the script, I made little adjustement to the original script, because the server I use to do my training create there own library to manage parallelisation called idr_torch, you can find the lines added in the main function :
import logging
import math
import os
import sys
from dataclasses import dataclass, field
from typing import Optional
from datasets import Dataset, load_dataset
import transformers
from transformers import (
CONFIG_MAPPING,
MODEL_FOR_MASKED_LM_MAPPING,
AutoConfig,
AutoModelForMaskedLM,
AutoTokenizer,
DataCollatorForWholeWordMask,
HfArgumentParser,
Trainer,
TrainingArguments,
set_seed,
)
from transformers.trainer_utils import get_last_checkpoint, is_main_process
import torch
from torch.distributed.elastic.multiprocessing.errors import record
import torch.distributed as dist
import idr_torch
logger = logging.getLogger(__name__)
MODEL_CONFIG_CLASSES = list(MODEL_FOR_MASKED_LM_MAPPING.keys())
MODEL_TYPES = tuple(conf.model_type for conf in MODEL_CONFIG_CLASSES)
@dataclass
class ModelArguments:
"""
Arguments pertaining to which model/config/tokenizer we are going to fine-tune, or train from scratch.
"""
model_name_or_path: Optional[str] = field(
default=None,
metadata={
"help": (
"The model checkpoint for weights initialization. Don't set if you want to train a model from scratch."
)
},
)
model_type: Optional[str] = field(
default=None,
metadata={"help": "If training from scratch, pass a model type from the list: " + ", ".join(MODEL_TYPES)},
)
config_overrides: Optional[str] = field(
default=None,
metadata={
"help": (
"Override some existing default config settings when a model is trained from scratch. Example: "
"n_embd=10,resid_pdrop=0.2,scale_attn_weights=false,summary_type=cls_index"
)
},
)
config_name: Optional[str] = field(
default=None, metadata={"help": "Pretrained config name or path if not the same as model_name"}
)
tokenizer_name: Optional[str] = field(
default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"}
)
cache_dir: Optional[str] = field(
default=None,
metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
)
use_fast_tokenizer: bool = field(
default=True,
metadata={"help": "Whether to use one of the fast tokenizer (backed by the tokenizers library) or not."},
)
model_revision: str = field(
default="main",
metadata={"help": "The specific model version to use (can be a branch name, tag name or commit id)."},
)
use_auth_token: bool = field(
default=False,
metadata={
"help": (
"Will use the token generated when running `huggingface-cli login` (necessary to use this script "
"with private models)."
)
},
)
def __post_init__(self):
if self.config_overrides is not None and (self.config_name is not None or self.model_name_or_path is not None):
raise ValueError(
"--config_overrides can't be used in combination with --config_name or --model_name_or_path"
)
@dataclass
class DataTrainingArguments:
"""
Arguments pertaining to what data we are going to input our model for training and eval.
"""
dataset_name: Optional[str] = field(
default=None, metadata={"help": "The name of the dataset to use (via the datasets library)."}
)
dataset_config_name: Optional[str] = field(
default=None, metadata={"help": "The configuration name of the dataset to use (via the datasets library)."}
)
train_file: Optional[str] = field(default=None, metadata={"help": "The input training data file (a text file)."})
validation_file: Optional[str] = field(
default=None,
metadata={"help": "An optional input evaluation data file to evaluate the perplexity on (a text file)."},
)
train_ref_file: Optional[str] = field(
default=None,
metadata={"help": "An optional input train ref data file for whole word masking in Chinese."},
)
validation_ref_file: Optional[str] = field(
default=None,
metadata={"help": "An optional input validation ref data file for whole word masking in Chinese."},
)
overwrite_cache: bool = field(
default=False, metadata={"help": "Overwrite the cached training and evaluation sets"}
)
validation_split_percentage: Optional[int] = field(
default=5,
metadata={
"help": "The percentage of the train set used as validation set in case there's no validation split"
},
)
max_seq_length: Optional[int] = field(
default=None,
metadata={
"help": (
"The maximum total input sequence length after tokenization. Sequences longer "
"than this will be truncated. Default to the max input length of the model."
)
},
)
preprocessing_num_workers: Optional[int] = field(
default=None,
metadata={"help": "The number of processes to use for the preprocessing."},
)
mlm_probability: float = field(
default=0.15, metadata={"help": "Ratio of tokens to mask for masked language modeling loss"}
)
pad_to_max_length: bool = field(
default=False,
metadata={
"help": (
"Whether to pad all samples to `max_seq_length`. "
"If False, will pad the samples dynamically when batching to the maximum length in the batch."
)
},
)
def __post_init__(self):
if self.train_file is not None:
extension = self.train_file.split(".")[-1]
assert extension in ["csv", "json", "txt"], "`train_file` should be a csv, a json or a txt file."
if self.validation_file is not None:
extension = self.validation_file.split(".")[-1]
assert extension in ["csv", "json", "txt"], "`validation_file` should be a csv, a json or a txt file."
def add_chinese_references(dataset, ref_file):
with open(ref_file, "r", encoding="utf-8") as f:
refs = [json.loads(line) for line in f.read().splitlines() if (len(line) > 0 and not line.isspace())]
assert len(dataset) == len(refs)
dataset_dict = {c: dataset[c] for c in dataset.column_names}
dataset_dict["chinese_ref"] = refs
return Dataset.from_dict(dataset_dict)
@record
def main():
os.environ['LOCAL_RANK'] = os.environ['SLURM_LOCALID']
dist.init_process_group(backend='nccl',
init_method='env://',
world_size=idr_torch.size,
rank=idr_torch.rank)
parser = HfArgumentParser((ModelArguments, DataTrainingArguments, TrainingArguments))
if len(sys.argv) == 2 and sys.argv[1].endswith(".json"):
# If we pass only one argument to the script and it's the path to a json file,
# let's parse it to get our arguments.
model_args, data_args, training_args = parser.parse_json_file(json_file=os.path.abspath(sys.argv[1]))
else:
model_args, data_args, training_args = parser.parse_args_into_dataclasses()
training_args.local_rank = idr_torch.local_rank
# 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:
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."
)
# Setup logging
logging.basicConfig(
format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
datefmt="%m/%d/%Y %H:%M:%S",
handlers=[logging.StreamHandler(sys.stdout)],
)
logger.setLevel(logging.INFO if is_main_process(training_args.local_rank) else logging.WARN)
# Log on each process the small summary:
logger.warning(
f"Process rank: {training_args.local_rank}, device: {training_args.device}, n_gpu: {training_args.n_gpu}"
+ f"distributed training: {bool(training_args.local_rank != -1)}, 16-bits training: {training_args.fp16}"
)
# Set the verbosity to info of the Transformers logger (on main process only):
if is_main_process(training_args.local_rank):
transformers.utils.logging.set_verbosity_info()
transformers.utils.logging.enable_default_handler()
transformers.utils.logging.enable_explicit_format()
logger.info("Training/evaluation parameters %s", training_args)
# Set seed before initializing model.
set_seed(training_args.seed)
# 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 load_dataset function guarantee that only one local process can concurrently
# download the dataset.
if data_args.dataset_name is not None:
# Downloading and loading a dataset from the hub.
datasets = load_dataset(data_args.dataset_name, data_args.dataset_config_name, cache_dir="/gpfswork/rech/khy/uvb95lb/scrip_continual_pretraining/Cache_mlm")
if "validation" not in datasets.keys():
datasets["validation"] = load_dataset(
data_args.dataset_name,
data_args.dataset_config_name,
split=f"train[:{data_args.validation_split_percentage}%]",
)
datasets["train"] = load_dataset(
data_args.dataset_name,
data_args.dataset_config_name,
split=f"train[{data_args.validation_split_percentage}%:]",
)
else:
data_files = {}
if data_args.train_file is not None:
data_files["train"] = data_args.train_file
extension = data_args.train_file.split(".")[-1]
if data_args.validation_file is not None:
data_files["validation"] = data_args.validation_file
extension = data_args.validation_file.split(".")[-1]
if extension == "txt":
extension = "text"
datasets = load_dataset(extension, data_files=data_files, cache_dir="/gpfswork/rech/khy/uvb95lb/scrip_continual_pretraining/Cache_mlm")
# 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.
# Load pretrained model and tokenizer
#
# Distributed training:
# The .from_pretrained methods guarantee that only one local process can concurrently
# download model & vocab.
config_kwargs = {
"cache_dir": model_args.cache_dir,
"revision": model_args.model_revision,
"use_auth_token": True if model_args.use_auth_token else None,
}
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,
"use_auth_token": True if model_args.use_auth_token else None,
}
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."
)
if model_args.model_name_or_path:
model = AutoModelForMaskedLM.from_pretrained(
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,
use_auth_token=True if model_args.use_auth_token else None,
local_files_only=True,
)
else:
logger.info("Training new model from scratch")
model = AutoModelForMaskedLM.from_config(config)
model.resize_token_embeddings(len(tokenizer))
# Preprocessing the datasets.
# First we tokenize all the texts.
if training_args.do_train:
column_names = datasets["train"].column_names
else:
column_names = datasets["validation"].column_names
text_column_name = "text" if "text" in column_names else column_names[0]
padding = "max_length" if data_args.pad_to_max_length else False
def tokenize_function(examples):
# Remove empty lines
examples["text"] = [line for line in examples["text"] if len(line) > 0 and not line.isspace()]
return tokenizer(examples["text"], padding=padding, truncation=True, max_length=data_args.max_seq_length)
tokenized_datasets = datasets.map(
tokenize_function,
batched=True,
num_proc=data_args.preprocessing_num_workers,
remove_columns=[text_column_name],
load_from_cache_file=not data_args.overwrite_cache
)
# Add the chinese references if provided
if data_args.train_ref_file is not None:
tokenized_datasets["train"] = add_chinese_references(tokenized_datasets["train"], data_args.train_ref_file)
if data_args.validation_ref_file is not None:
tokenized_datasets["validation"] = add_chinese_references(
tokenized_datasets["validation"], data_args.validation_ref_file
)
# If we have ref files, need to avoid it removed by trainer
has_ref = data_args.train_ref_file or data_args.validation_ref_file
if has_ref:
training_args.remove_unused_columns = False
# Data collator
# This one will take care of randomly masking the tokens.
data_collator = DataCollatorForWholeWordMask(tokenizer=tokenizer, mlm_probability=data_args.mlm_probability)
# Initialize our Trainer
trainer = Trainer(
model=model,
args=training_args,
train_dataset=tokenized_datasets["train"] if training_args.do_train else None,
eval_dataset=tokenized_datasets["validation"] if training_args.do_eval else None,
tokenizer=tokenizer,
data_collator=data_collator,
)
# Training
if training_args.do_train:
if last_checkpoint is not None:
checkpoint = last_checkpoint
elif model_args.model_name_or_path is not None and os.path.isdir(model_args.model_name_or_path):
checkpoint = model_args.model_name_or_path
else:
checkpoint = None
train_result = trainer.train(resume_from_checkpoint=checkpoint)
trainer.save_model() # Saves the tokenizer too for easy upload
output_train_file = os.path.join(training_args.output_dir, "train_results.txt")
if trainer.is_world_process_zero():
with open(output_train_file, "w") as writer:
logger.info("***** Train results *****")
for key, value in sorted(train_result.metrics.items()):
logger.info(f" {key} = {value}")
writer.write(f"{key} = {value}\n")
# Need to save the state, since Trainer.save_model saves only the tokenizer with the model
trainer.state.save_to_json(os.path.join(training_args.output_dir, "trainer_state.json"))
# Evaluation
results = {}
if training_args.do_eval:
logger.info("*** Evaluate ***")
eval_output = trainer.evaluate()
perplexity = math.exp(eval_output["eval_loss"])
results["perplexity"] = perplexity
output_eval_file = os.path.join(training_args.output_dir, "eval_results_mlm_wwm.txt")
if trainer.is_world_process_zero():
with open(output_eval_file, "w") as writer:
logger.info("***** Eval results *****")
for key, value in sorted(results.items()):
logger.info(f" {key} = {value}")
writer.write(f"{key} = {value}\n")
return results
def _mp_fn(index):
# For xla_spawn (TPUs)
main()
if __name__ == "__main__":
main()```Gentle ping @muellerzr
System Info
Hi,
I am currently trying to use the script run_mlm_wwm.py to perform a continual pretrianing on the Whole Word mazking task, on a Bert model, my problem occured when I am trying to use multpiles GPU, when the number of GPU force my server to use severals nodes then I get an error. I think there's a locking problem linked to the parallel file system when I go over several nodes, I would have to set a different cache for each process (for example by adding the global rank) or for each process on the same node but i don't succeed to do it so far
Is there the error message i get :
`Loading pytorch-gpu/py3/2.1.1 Loading requirement: cuda/11.8.0 nccl/2.18.5-1-cuda cudnn/8.7.0.84-cuda gcc/8.5.0 openmpi/4.1.5-cuda intel-mkl/2020.4 magma/2.7.1-cuda sox/14.4.2 sparsehash/2.0.3 libjpeg-turbo/2.1.3 ffmpeg/4.4.4
COMET_API_KEYis not set. 6: comet_ml is installed butCOMET_API_KEYis not set. 7: comet_ml is installed butCOMET_API_KEYis not set. 8: comet_ml is installed butCOMET_API_KEYis not set. 9: comet_ml is installed butCOMET_API_KEYis not set. 10: comet_ml is installed butCOMET_API_KEYis not set. 11: comet_ml is installed butCOMET_API_KEYis not set. 12: comet_ml is installed butCOMET_API_KEYis not set. 13: comet_ml is installed butCOMET_API_KEYis not set. 14: comet_ml is installed butCOMET_API_KEYis not set. 15: comet_ml is installed butCOMET_API_KEYis not set. 16: comet_ml is installed butCOMET_API_KEYis not set. 17: comet_ml is installed butCOMET_API_KEYis not set. 3: comet_ml is installed butCOMET_API_KEYis not set. 4: comet_ml is installed butCOMET_API_KEYis not set. 5: comet_ml is installed butCOMET_API_KEYis not set. 1: comet_ml is installed butCOMET_API_KEYis not set. 2: comet_ml is installed butCOMET_API_KEYis not set. 2: 05/04/2024 06:14:12 - WARNING - main - Process rank: 2, device: cuda:2, n_gpu: 1distributed training: True, 16-bits training: True 1: 05/04/2024 06:14:12 - WARNING - main - Process rank: 1, device: cuda:1, n_gpu: 1distributed training: True, 16-bits training: True 3: 05/04/2024 06:14:12 - WARNING - main - Process rank: 3, device: cuda:3, n_gpu: 1distributed training: True, 16-bits training: True 4: 05/04/2024 06:14:12 - WARNING - main - Process rank: 4, device: cuda:4, n_gpu: 1distributed training: True, 16-bits training: True 5: 05/04/2024 06:14:12 - WARNING - main - Process rank: 5, device: cuda:5, n_gpu: 1distributed training: True, 16-bits training: True 14: 05/04/2024 06:14:12 - WARNING - main - Process rank: 2, device: cuda:2, n_gpu: 1distributed training: True, 16-bits training: True 15: 05/04/2024 06:14:12 - WARNING - main - Process rank: 3, device: cuda:3, n_gpu: 1distributed training: True, 16-bits training: True 16: 05/04/2024 06:14:12 - WARNING - main - Process rank: 4, device: cuda:4, n_gpu: 1distributed training: True, 16-bits training: True 12: 05/04/2024 06:14:12 - WARNING - main - Process rank: 0, device: cuda:0, n_gpu: 1distributed training: True, 16-bits training: True 13: 05/04/2024 06:14:12 - WARNING - main - Process rank: 1, device: cuda:1, n_gpu: 1distributed training: True, 16-bits training: True 17: 05/04/2024 06:14:12 - WARNING - main - Process rank: 5, device: cuda:5, n_gpu: 1distributed training: True, 16-bits training: True 12: 05/04/2024 06:14:12 - INFO - main - Training/evaluation parameters TrainingArguments( 12: _n_gpu=1, 12: adafactor=False, 12: adam_beta1=0.9, 12: adam_beta2=0.999, 12: adam_epsilon=1e-08, 12: auto_find_batch_size=False, 12: bf16=False, 12: bf16_full_eval=False, 12: data_seed=None, 12: dataloader_drop_last=False, 12: dataloader_num_workers=0, 12: dataloader_pin_memory=True, 12: ddp_backend=None, 12: ddp_broadcast_buffers=None, 12: ddp_bucket_cap_mb=None, 12: ddp_find_unused_parameters=False, 12: ddp_timeout=600, 12: debug=[], 12: deepspeed=None, 12: disable_tqdm=False, 12: dispatch_batches=None, 12: do_eval=True, 12: do_predict=False, 12: do_train=True, 12: eval_accumulation_steps=None, 12: eval_delay=0, 12: eval_steps=None, 12: evaluation_strategy=IntervalStrategy.NO, 12: fp16=True, 12: fp16_backend=auto, 12: fp16_full_eval=False, 12: fp16_opt_level=O1, 12: fsdp=[], 12: fsdp_config={'min_num_params': 0, 'xla': False, 'xla_fsdp_grad_ckpt': False}, 12: fsdp_min_num_params=0, 12: fsdp_transformer_layer_cls_to_wrap=None, 12: full_determinism=False, 12: gradient_accumulation_steps=1, 12: gradient_checkpointing=False, 12: gradient_checkpointing_kwargs=None, 12: greater_is_better=None, 12: group_by_length=False, 12: half_precision_backend=auto, 12: hub_always_push=False, 12: hub_model_id=None, 12: hub_private_repo=False, 12: hub_strategy=HubStrategy.EVERY_SAVE, 12: hub_token=Who can help?
No response
Information
Tasks
examplesfolder (such as GLUE/SQuAD, ...)Reproduction
14: OSError: [Errno 39] Directory not empty: '/scrip_continual_pretraining/Cache_mlm/text/default-d0870639fca1403e/0.0.0/c4a140d10f020282918b5dd1b8a49f0104729c6177f60a6b49ec2a365ec69f34.incomplete' Generating train split: 20130863 examples [00:54, 368921.36 examples/s] 9: Traceback (most recent call last): 9: File "/gpfsdswork/projects/rech/khy/uvb95lb/scrip_continual_pretraining/run_mlm_wwm.py", line 450, in
9: main()
9: File "/gpfslocalsup/pub/anaconda-py3/2023.09/envs/pytorch-gpu-2.1.1+py3.11.5/lib/python3.11/site-packages/torch/distributed/elastic/multiprocessing/errors/init.py", line 346, in wrapper
9: return f(args, kwargs)
9: ^^^^^^^^^^^^^^^^^^
9: File "/scrip_continual_pretraining/run_mlm_wwm.py", line 294, in main
9: datasets = load_dataset(extension, data_files=data_files, cache_dir="/scrip_continual_pretraining/Cache_mlm")
9: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9: File "/gpfslocalsup/pub/anaconda-py3/2023.09/envs/pytorch-gpu-2.1.1+py3.11.5/lib/python3.11/site-packages/datasets/load.py", line 2152, in load_dataset
9: builder_instance.download_and_prepare(
9: File "/gpfslocalsup/pub/anaconda-py3/2023.09/envs/pytorch-gpu-2.1.1+py3.11.5/lib/python3.11/site-packages/datasets/builder.py", line 948, in download_and_prepare
9: self._download_and_prepare(
9: File "/gpfslocalsup/pub/anaconda-py3/2023.09/envs/pytorch-gpu-2.1.1+py3.11.5/lib/python3.11/site-packages/datasets/builder.py", line 1045, in _download_and_prepare
9: raise OSError(
9: OSError: Cannot find data file.
9: Original error:
9: [Errno 2] No such file or directory: '/scrip_continual_pretraining/Cache_mlm/text/default-d0870639fca1403e/0.0.0/c4a140d10f020282918b5dd1b8a49f0104729c6177f60a6b49ec2a365ec69f34.incomplete/text-train-00000-00001-of-NNNNN.arrow'
Downloading data files: 100%|██████████| 2/2 [00:00<00:00, 17962.76it/s]
Extracting data files: 100%|██████████| 2/2 [00:00<00:00, 59.23it/s]
Generating validation split: 3246757 examples [00:07, 410442.68 examples/s]
srun: error: jean-zay-iam07: task 14: Exited with exit code 1
srun: Terminating StepId=1741717.0
0: slurmstepd: error: STEP 1741717.0 ON jean-zay-iam05 CANCELLED AT 2024-05-04T06:15:08 ***
Generating train split: 122206 examples [00:00, 396783.29 examples/s]
2: split: 3308013 examples [00:07, 424316.76 examples/s]
srun: error: jean-zay-iam07: tasks 12-13,15-16: Terminated
srun: error: jean-zay-iam05: tasks 0-2,4-5: Terminated
Expected behavior
I expect that my training succed to run at least one epoch