Hello all !
I am trying to fine-tune a Bert2Bert Model for the translation task, using deepspeed and accelerate.
I am following the suggested post and the examples/pytorch/translation both by Hugginface.
Unfortunately, whwn I have to generate the translation after the training I am having the same word reapeated 20 time. If i change hyperparameters I have the same constant value of
{'bleu': 0.09453580071770594} no metter what I change.
Do you know if can be something in the code that is wrong or it is a problem of the training?
I am using:
--learning_rate 5e-5 --num_train_epochs 3 --source_lang source --target_lang target--checkpointing_steps epoch --with_tracking yes --source_prefix translation
and my dataset is in the format:
{"translation": {"source": "sentence", "target": "sentence"}}
Thank you very much ! I have no more ideas…
...
parser.add_argument(
"--num_beams",
type=int,
default=None,
help=(
"Number of beams to use for evaluation. This argument will be "
"passed to ``model.generate``, which is used during ``evaluate`` and ``predict``."
),
)
parser.add_argument(
"--max_source_length",
type=int,
default=1024,
help=(
"The maximum total input sequence length after "
"tokenization.Sequences longer than this will be truncated, sequences shorter will be padded."
),
)
parser.add_argument(
"--max_target_length",
type=int,
default=512,
help=(
"The maximum total sequence length for target text after "
"tokenization. Sequences longer than this will be truncated, sequences shorter will be padded."
"during ``evaluate`` and ``predict``."
),
)
parser.add_argument(
"--val_max_target_length",
type=int,
default=None,
help=(
"The maximum total sequence length for validation "
"target text after tokenization.Sequences longer than this will be truncated, sequences shorter will be "
"padded. Will default to `max_target_length`.This argument is also used to override the ``max_length`` "
"param of ``model.generate``, which is used during ``evaluate`` and ``predict``."
),
)
parser.add_argument(
"--pad_to_max_length",
type=bool,
default=False,
help=(
"Whether to pad all samples to model maximum sentence "
"length. If False, will pad the samples dynamically when batching to the maximum length in the batch. More"
"efficient on GPU but very bad for TPU."
),
)
parser.add_argument(
"--validation_file", type=str, default=None, help="A csv or a json file containing the validation data."
)
parser.add_argument(
"--ignore_pad_token_for_loss",
type=bool,
default=True,
help="Whether to ignore the tokens corresponding to padded labels in the loss computation or not.",
)
parser.add_argument("--source_lang", type=str, default=None, help="Source language id for translation.")
parser.add_argument("--target_lang", type=str, default=None, help="Target language id for translation.")
parser.add_argument(
"--source_prefix",
type=str,
default=None,
help="A prefix to add before every source text (useful for T5 models).",
)
parser.add_argument(
"--preprocessing_num_workers",
type=int,
default=None,
help="The number of processes to use for the preprocessing.",
)
parser.add_argument(
"--overwrite_cache", type=bool, default=None, help="Overwrite the cached training and evaluation sets"
)
parser.add_argument(
"--max_length",
type=int,
default=512,
help=(
"The maximum total input sequence length after tokenization. Sequences longer than this will be truncated,"
" sequences shorter will be padded if `--pad_to_max_lengh` is passed."
),
)
...
parser.add_argument("--weight_decay", type=float, default=0.0, help="Weight decay to use.")
parser.add_argument("--num_train_epochs", type=int, default=3, help="Total number of training epochs to perform.")
...
parser.add_argument(
"--gradient_accumulation_steps",
type=int,
default=1,
help="Number of updates steps to accumulate before performing a backward/update pass.",
)
parser.add_argument(
"--lr_scheduler_type",
type=SchedulerType,
default="linear",
help="The scheduler type to use.",
choices=["linear", "cosine", "cosine_with_restarts", "polynomial", "constant", "constant_with_warmup"],
)
parser.add_argument(
"--num_warmup_steps", type=int, default=0, help="Number of steps for the warmup in the lr scheduler."
)
...
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.",
)
# Whether to load the best model at the end of training
parser.add_argument(
"--load_best_model",
action="store_true",
help="Whether to load the best model at the end of training",
)
parser.add_argument(
"--logging_steps",
type=int,
default=None,
help="log every n steps",
)
parser.add_argument(
"--with_tracking",
#action="store_true",
help="Whether to enable experiment trackers for logging.",
)
parser.add_argument(
"--report_to",
type=str,
default="all",
help=(
'The integration to report the results and logs to. Supported platforms are `"tensorboard"`,'
' `"wandb"` and `"comet_ml"`. Use `"all"` (default) to report to all integrations.'
"Only applicable when `--with_tracking` is passed."
),
)
...
args = parser.parse_args()
if args.dataset_name is None and args.train_file is None and args.validation_file is None:
raise ValueError("Need either a task name or a training/validation file.")
if args.train_file is not None:
extension = args.train_file.split(".")[-1]
assert extension in ["csv", "json"], "`train_file` should be a csv or a json file."
if args.validation_file is not None:
extension = args.validation_file.split(".")[-1]
assert extension in ["csv", "json"], "`validation_file` should be a csv or a json file."
...
return args
def main():
# Parse the arguments
args = parse_args()
...
accelerator = (
Accelerator(log_with=args.report_to, logging_dir=args.output_dir) if args.with_tracking else Accelerator()
)
# Make one log on every process with the configuration for debugging.
logging.basicConfig(
format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
datefmt="%m/%d/%Y %H:%M:%S",
level=logging.INFO,
)
logger.info(accelerator.state, main_process_only=False)
if accelerator.is_local_main_process:
datasets.utils.logging.set_verbosity_warning()
transformers.utils.logging.set_verbosity_info()
else:
datasets.utils.logging.set_verbosity_error()
transformers.utils.logging.set_verbosity_error()
...
# Handle the repository creation
if accelerator.is_main_process:
if args.push_to_hub:
if args.hub_model_id is None:
repo_name = get_full_repo_name(Path(args.output_dir).name, token=args.hub_token)
else:
repo_name = args.hub_model_id
repo = Repository(args.output_dir, clone_from=repo_name)
with open(os.path.join(args.output_dir, ".gitignore"), "w+") as gitignore:
if "step_*" not in gitignore:
gitignore.write("step_*\n")
if "epoch_*" not in gitignore:
gitignore.write("epoch_*\n")
elif args.output_dir is not None:
os.makedirs(args.output_dir, exist_ok=True)
accelerator.wait_for_everyone()
# Get the datasets: you can either provide your own CSV/JSON/TXT training and evaluation files (see below)
# 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 args.dataset_name is not None:
# Downloading and loading a dataset from the hub.
raw_datasets = load_dataset(args.dataset_name, args.dataset_config_name)
else:
data_files = {}
if args.train_file is not None:
data_files["train"] = args.train_file
if args.validation_file is not None:
data_files["validation"] = args.validation_file
extension = args.train_file.split(".")[-1]
raw_datasets = load_dataset(extension, data_files=data_files)
print(raw_datasets)
# 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.html.
# Load pretrained model and tokenizer
#
# In distributed training, the .from_pretrained methods guarantee that only one local process can concurrently
# download model & vocab.
if args.config_name:
config = AutoConfig.from_pretrained(args.config_name)
elif args.model_name_or_path:
config = AutoConfig.from_pretrained(args.model_name_or_path)
else:
config = CONFIG_MAPPING[args.model_type]()
logger.warning("You are instantiating a new config instance from scratch.")
if args.tokenizer_name:
tokenizer = AutoTokenizer.from_pretrained(args.tokenizer_name, use_fast=not args.use_slow_tokenizer)
elif args.model_name_or_path:
tokenizer = AutoTokenizer.from_pretrained(args.model_name_or_path, use_fast=not args.use_slow_tokenizer)
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 args.model_name_or_path:
vocabsize = 30
max_length = 512
encoder_config = BertConfig(vocab_size = vocabsize,
max_position_embeddings = max_length+64, # this shuold be some large value
num_attention_heads = 16,
max_length = 512,
num_hidden_layers = 30,
hidden_size = 1024,
type_vocab_size = 1,
).from_pretrained(args.model_name_or_path)
encoder = BertModel(config=encoder_config)
decoder_config = BertConfig(vocab_size = vocabsize,
max_position_embeddings = max_length+64, # this shuold be some large value
num_attention_heads = 16,
max_length = 512,
num_hidden_layers = 30,
hidden_size = 1024,
type_vocab_size = 1,
is_decoder=True,
add_cross_attention=True,
).from_pretrained(args.model_name_or_path) # Very Important
# Define encoder decoder model
decoder = BertForMaskedLM(config=decoder_config)
# Define encoder decoder model
config = EncoderDecoderConfig.from_encoder_decoder_configs(encoder_config, decoder_config)
model = EncoderDecoderModel(config=config)
model.config.decoder_start_token_id = tokenizer.cls_token_id
model.config.eos_token_id = tokenizer.sep_token_id
model.config.pad_token_id = tokenizer.pad_token_id
model.config.vocab_size = model.config.encoder.vocab_size
else:
logger.info("Training new model from scratch")
model = AutoModelForSeq2SeqLM.from_config(config)
#model.resize_token_embeddings(len(tokenizer))
# Set decoder_start_token_id
if model.config.decoder_start_token_id is None and isinstance(tokenizer, (MBartTokenizer, MBartTokenizerFast)):
assert (
args.target_lang is not None and args.source_lang is not None
), "mBart requires --target_lang and --source_lang"
if isinstance(tokenizer, MBartTokenizer):
model.config.decoder_start_token_id = tokenizer.lang_code_to_id[args.target_lang]
else:
model.config.decoder_start_token_id = tokenizer.convert_tokens_to_ids(args.target_lang)
if model.config.decoder_start_token_id is None:
#model.config.decoder_start_token_id = tokenizer.convert_tokens_to_ids(args.target_lang)
raise ValueError("Make sure that `config.decoder_start_token_id` is correctly defined")
prefix = args.source_prefix if args.source_prefix is not None else ""
# Preprocessing the datasets.
# First we tokenize all the texts.
column_names = raw_datasets["train"].column_names
# For translation we set the codes of our source and target languages (only useful for mBART, the others will
# ignore those attributes).
if isinstance(tokenizer, (MBartTokenizer, MBartTokenizerFast)):
if args.source_lang is not None:
tokenizer.src_lang = args.source_lang
if args.target_lang is not None:
tokenizer.tgt_lang = args.target_lang
# Get the language codes for input/target.
source_lang = args.source_lang.split("_")[0]
target_lang = args.target_lang.split("_")[0]
padding = "max_length" if args.pad_to_max_length else False
# Temporarily set max_target_length for training.
max_target_length = args.max_target_length
padding = "max_length" if args.pad_to_max_length else False
def preprocess_function(examples):
inputs = [ex[source_lang] for ex in examples["translation"]]
targets = [ex[target_lang] for ex in examples["translation"]]
inputs = [prefix + inp for inp in inputs]
model_inputs = tokenizer(inputs, max_length=args.max_source_length, padding=padding, truncation=True)
# Tokenize targets with the `text_target` keyword argument
#labels = tokenizer(text_target=targets, max_length=max_target_length, padding=padding, truncation=True)
labels = tokenizer(targets, max_length=max_target_length, padding=padding, truncation=True)
# If we are padding here, replace all tokenizer.pad_token_id in the labels by -100 when we want to ignore
# padding in the loss.
if padding == "max_length" and args.ignore_pad_token_for_loss:
labels["input_ids"] = [
[(l if l != tokenizer.pad_token_id else -100) for l in label] for label in labels["input_ids"]
]
model_inputs["labels"] = labels["input_ids"]
return model_inputs
with accelerator.main_process_first():
processed_datasets = raw_datasets.map(
preprocess_function,
batched=True,
num_proc=args.preprocessing_num_workers,
remove_columns=column_names,
load_from_cache_file=not args.overwrite_cache,
desc="Running tokenizer on dataset",
)
train_dataset = processed_datasets["train"]
eval_dataset = processed_datasets["validation"]
# Log a few random samples from the training set:
for index in random.sample(range(len(train_dataset)), 3):
logger.info(f"Sample {index} of the training set: {train_dataset[index]}.")
# DataLoaders creation:
label_pad_token_id = -100 if args.ignore_pad_token_for_loss else tokenizer.pad_token_id
if args.pad_to_max_length:
# If padding was already done ot max length, we use the default data collator that will just convert everything
# to tensors.
data_collator = default_data_collator
else:
# Otherwise, `DataCollatorWithPadding` will apply dynamic padding for us (by padding to the maximum length of
# the samples passed). When using mixed precision, we add `pad_to_multiple_of=8` to pad all tensors to multiple
# of 8s, which will enable the use of Tensor Cores on NVIDIA hardware with compute capability >= 7.5 (Volta).
data_collator = DataCollatorForSeq2Seq(
tokenizer,
model=model,
label_pad_token_id=label_pad_token_id,
pad_to_multiple_of=8 if accelerator.use_fp16 else None,
)
train_dataloader = DataLoader(
train_dataset, shuffle=True, collate_fn=data_collator, batch_size=args.per_device_train_batch_size
)
eval_dataloader = DataLoader(eval_dataset, collate_fn=data_collator, batch_size=args.per_device_eval_batch_size)
# Optimizer
# Split weights in two groups, one with weight decay and the other not.
no_decay = ["bias", "LayerNorm.weight"]
optimizer_grouped_parameters = [
{
"params": [p for n, p in model.named_parameters() if not any(nd in n for nd in no_decay)],
"weight_decay": args.weight_decay,
},
{
"params": [p for n, p in model.named_parameters() if any(nd in n for nd in no_decay)],
"weight_decay": 0.0,
},
]
# Creates Dummy Optimizer if `optimizer` was spcified in the config file else creates Adam Optimizer
optimizer_cls = (
torch.optim.Adam
if accelerator.state.deepspeed_plugin is None
or "optimizer" not in accelerator.state.deepspeed_plugin.deepspeed_config
else DummyOptim
)
optimizer = optimizer_cls(optimizer_grouped_parameters, lr=args.learning_rate)
# Get gradient accumulation steps from deepspeed config if available
if accelerator.state.deepspeed_plugin is not None:
args.gradient_accumulation_steps = accelerator.state.deepspeed_plugin.deepspeed_config[
"gradient_accumulation_steps"
]
#optimizer = torch.optim.AdamW(optimizer_grouped_parameters, lr=args.learning_rate)
# Scheduler and math around the number of training steps.
overrode_max_train_steps = False
num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
if args.max_train_steps is None:
args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
overrode_max_train_steps = True
# Creates Dummy Scheduler if `scheduler` was spcified in the config file else creates `args.lr_scheduler_type` Scheduler
if (
accelerator.state.deepspeed_plugin is None
or "scheduler" not in accelerator.state.deepspeed_plugin.deepspeed_config
):
lr_scheduler = get_scheduler(
name=args.lr_scheduler_type,
optimizer=optimizer,
num_warmup_steps=args.num_warmup_steps,
num_training_steps=args.max_train_steps,
)
else:
lr_scheduler = DummyScheduler(
optimizer, total_num_steps=args.max_train_steps, warmup_num_steps=args.num_warmup_steps
)
# Prepare everything with our `accelerator`.
model, optimizer, train_dataloader, eval_dataloader, lr_scheduler = accelerator.prepare(
model, optimizer, train_dataloader, eval_dataloader, lr_scheduler
)
# We need to recalculate our total training steps as the size of the training dataloader may have changed.
num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
if overrode_max_train_steps:
args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
# Afterwards we recalculate our number of training epochs
args.num_train_epochs = math.ceil(args.max_train_steps / num_update_steps_per_epoch)
# Figure out how many steps we should save the Accelerator states
if hasattr(args.checkpointing_steps, "isdigit"):
checkpointing_steps = args.checkpointing_steps
if args.checkpointing_steps.isdigit():
checkpointing_steps = int(args.checkpointing_steps)
else:
checkpointing_steps = None
# We need to initialize the trackers we use, and also store our configuration.
# We initialize the trackers only on main process because `accelerator.log`
# only logs on main process and we don't want empty logs/runs on other processes.
if args.with_tracking:
if accelerator.is_main_process:
experiment_config = vars(args)
# TensorBoard cannot log Enums, need the raw value
experiment_config["lr_scheduler_type"] = experiment_config["lr_scheduler_type"].value
accelerator.init_trackers(args.report_name, experiment_config)
metric = load_metric("sacrebleu")
def postprocess_text(preds, labels):
preds = [pred.strip() for pred in preds]
labels = [[label.strip()] for label in labels]
return preds, labels
# Train!
total_batch_size = args.per_device_train_batch_size * accelerator.num_processes * args.gradient_accumulation_steps
logger.info("***** Running training *****")
logger.info(f" Num examples = {len(train_dataset)}")
logger.info(f" Num Epochs = {args.num_train_epochs}")
logger.info(f" Instantaneous batch size per device = {args.per_device_train_batch_size}")
logger.info(f" Total train batch size (w. parallel, distributed & accumulation) = {total_batch_size}")
logger.info(f" Gradient Accumulation steps = {args.gradient_accumulation_steps}")
logger.info(f" Total optimization steps = {args.max_train_steps}")
# Only show the progress bar once on each machine.
progress_bar = tqdm(range(args.max_train_steps), disable=not accelerator.is_local_main_process)
#progress_bar = tqdm(range(args.max_train_steps))
completed_steps = 0
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)
for epoch in range(starting_epoch, args.num_train_epochs):
start_time = time()
print('--Start training loop...')
print('Epoch',epoch)
print('accelerator.is_main_process',accelerator.is_main_process)
model.train()
if args.with_tracking:
total_loss = 0
for step, batch in enumerate(train_dataloader):
torch.cuda.empty_cache()
# We need to skip steps until we reach the resumed step
if args.resume_from_checkpoint and epoch == starting_epoch:
if resume_step is not None and step < resume_step:
completed_steps += 1
continue
outputs = model(**batch)
#outputs = model(input_ids=batch["input_ids"],decoder_input_ids=batch["input_ids"],labels=batch["input_ids"])
loss = outputs.loss
# We keep track of the loss at each epoch
if args.with_tracking:
total_loss += loss.detach().float()
loss = loss / args.gradient_accumulation_steps
accelerator.backward(loss)
if step % args.gradient_accumulation_steps == 0 or step == len(train_dataloader) - 1:
optimizer.step()
lr_scheduler.step()
optimizer.zero_grad()
progress_bar.update(1)
completed_steps += 1
if isinstance(args.logging_steps, int):
if completed_steps % args.logging_steps == 0:
steps_this_epoch = completed_steps % len(train_dataloader)
train_loss = total_loss.item() / steps_this_epoch
#train_perplexity = math.exp(train_loss)
accelerator.log(
{
"train_loss": train_loss,
#"train_perplexity": train_perplexity,
"epoch": epoch,
"step": completed_steps,
"steps_this_epoch": steps_this_epoch,
},
step=completed_steps,
)
logger.info(
f"Epoch: {epoch}, Step: {completed_steps}, Loss: {train_loss}, Perplexity: {train_perplexity}"
)
if isinstance(checkpointing_steps, int):
if completed_steps % checkpointing_steps == 0:
output_dir = f"step_{completed_steps }"
if args.output_dir is not None:
output_dir = os.path.join(args.output_dir, output_dir)
accelerator.save_state(output_dir)
if completed_steps >= args.max_train_steps:
break
end_time = time()
logger.info(f"Epoch {epoch} training took {end_time-start_time} seconds")
print('Completed steps',completed_steps)
print('Max train steps',args.max_train_steps)
print('Epoch',epoch)
print('--Starting evaluation...')
model.eval()
if args.val_max_target_length is None:
args.val_max_target_length = args.max_target_length
gen_kwargs = {
"max_length": args.val_max_target_length if args is not None else config.max_length,
"num_beams": args.num_beams,
}
samples_seen = 0
eval_batch_counter=0
for step, batch in enumerate(eval_dataloader):
eval_batch_counter += 1
print(eval_batch_counter)
print('New batch eval loop')
torch.cuda.empty_cache()
with torch.no_grad():
generated_tokens = accelerator.unwrap_model(model).generate(
batch["input_ids"],
attention_mask=batch["attention_mask"],
**gen_kwargs,
)
generated_tokens = accelerator.pad_across_processes(
generated_tokens, dim=1, pad_index=tokenizer.pad_token_id
)
labels = batch["labels"]
if not args.pad_to_max_length:
# If we did not pad to max length, we need to pad the labels too
labels = accelerator.pad_across_processes(batch["labels"], dim=1, pad_index=tokenizer.pad_token_id)
generated_tokens = accelerator.gather(generated_tokens).cpu().numpy()
labels = accelerator.gather(labels).cpu().numpy()
if args.ignore_pad_token_for_loss:
# Replace -100 in the labels as we can't decode them.
labels = np.where(labels != -100, labels, tokenizer.pad_token_id)
decoded_preds = tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)
decoded_labels = tokenizer.batch_decode(labels, skip_special_tokens=True)
decoded_preds, decoded_labels = postprocess_text(decoded_preds, decoded_labels)
# If we are in a multiprocess environment, the last batch has duplicates
if accelerator.num_processes > 1:
if step == len(eval_dataloader) - 1:
decoded_preds = decoded_preds[: len(eval_dataloader.dataset) - samples_seen]
decoded_labels = decoded_labels[: len(eval_dataloader.dataset) - samples_seen]
else:
samples_seen += len(decoded_labels)
metric.add_batch(predictions=decoded_preds, references=decoded_labels)
print('End val loop')
print('--Computing metrics...')
eval_metric = metric.compute()
logger.info({"bleu": eval_metric["score"]})
if args.with_tracking:
accelerator.log(
{
#"blue": eval_metric["score"],
"train_loss": total_loss.item() / len(train_dataloader),
#"train_perplexity" : math.exp(total_loss),
"epoch": epoch,
#"step": completed_steps,
},
step=completed_steps,
)
if args.push_to_hub and epoch < args.num_train_epochs - 1:
accelerator.wait_for_everyone()
unwrapped_model = accelerator.unwrap_model(model)
unwrapped_model.save_pretrained(
args.output_dir, is_main_process=accelerator.is_main_process, save_function=accelerator.save
)
if accelerator.is_main_process:
tokenizer.save_pretrained(args.output_dir)
repo.push_to_hub(
commit_message=f"Training in progress epoch {epoch}", blocking=False, auto_lfs_prune=True
)
if args.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)
print('End one all training loop')
if args.output_dir is not None:
accelerator.wait_for_everyone()
unwrapped_model = accelerator.unwrap_model(model)
unwrapped_model.save_pretrained(
args.output_dir,
is_main_process=accelerator.is_main_process, save_function=accelerator.save,state_dict=accelerator.get_state_dict(model)
)
if accelerator.is_main_process:
tokenizer.save_pretrained(args.output_dir)
if args.push_to_hub:
repo.push_to_hub(commit_message="End of training", auto_lfs_prune=True)
with open(os.path.join(args.output_dir, "all_results.json"), "w") as f:
json.dump({"eval_bleu": eval_metric["score"]}, f)
if __name__ == "__main__":
main()
My generation step is the following:
model = EncoderDecoderModel.from_pretrained("./training_out")
tokenizer = BertTokenizer.from_pretrained("./training_out")
inputs = tokenizer(sentence,return_tensors="pt",)
outputs = model.generate(inputs["input_ids"], attention_mask=inputs['attention_mask'],decoder_start_token_id = tokenizer.cls_token_id)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))