Sizes of tensors must match except in dimension 0

Beginners
1

I’m trying to pre-train BERT with my dataset using Trainer and torchrun. It works on 1 GPU, fails on 2 GPUs. I’m running with torchrun --nproc-per-node 2 my_benchmark.py

RuntimeError: Sizes of tensors must match except in dimension 0. Expected size 472 but got size 309 for tensor number 1 in the list.

472 is the length of an encoded sample. Why would there be a tensor of length 309 only when I run with 2 GPUs?

Here is an edited backtrace:

File "baseline_pytorch.py", line 112, in run
  trainer.train()
File "site-packages/transformers/trainer.py", line 1553, in train
  return inner_training_loop(
         ^^^^^^^^^^^^^^^^^^^^
File "site-packages/transformers/trainer.py", line 1813, in _inner_training_loop
  for step, inputs in enumerate(epoch_iterator):
File "site-packages/accelerate/data_loader.py", line 560, in __iter__
  next_batch, next_batch_info = self._fetch_batches(main_iterator)
                                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "site-packages/accelerate/data_loader.py", line 524, in _fetch_batches
  batch = concatenate(batches, dim=0)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "site-packages/accelerate/utils/operations.py", line 496, in concatenate
  return type(data[0])({k: concatenate([d[k] for d in data], dim=dim) for k in data[0].keys()})
                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "site-packages/accelerate/utils/operations.py", line 496, in <dictcomp>
  return type(data[0])({k: concatenate([d[k] for d in data], dim=dim) for k in data[0].keys()})
                           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "site-packages/accelerate/utils/operations.py", line 499, in concatenate
  return torch.cat(data, dim=dim)
         ^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: Sizes of tensors must match except in dimension 0. Expected size 472 but got size 309 for tensor number 1 in the list.

And some training args:

TrainingArguments(
_n_gpu=1,
adafactor=False,
adam_beta1=0.9,
adam_beta2=0.98,
adam_epsilon=2e-08,
auto_find_batch_size=False,
bf16=False,
bf16_full_eval=False,
data_seed=None,
dataloader_drop_last=False,
dataloader_num_workers=0,
dataloader_pin_memory=True,
ddp_backend=None,
ddp_broadcast_buffers=None,
ddp_bucket_cap_mb=None,
ddp_find_unused_parameters=None,
ddp_timeout=1800,
debug=[],
deepspeed=None,
disable_tqdm=False,
dispatch_batches=None,
do_eval=False,
do_predict=False,
do_train=True,
eval_accumulation_steps=None,
eval_delay=0,
eval_steps=None,
evaluation_strategy=IntervalStrategy.NO,
fp16=False,
fp16_backend=auto,
fp16_full_eval=False,
fp16_opt_level=O1,
fsdp=[],
fsdp_config={'min_num_params': 0, 'xla': False, 'xla_fsdp_grad_ckpt': False},
fsdp_min_num_params=0,
fsdp_transformer_layer_cls_to_wrap=None,
full_determinism=False,
gradient_accumulation_steps=1,
gradient_checkpointing=False,
greater_is_better=None,
group_by_length=False,
half_precision_backend=auto,
hub_always_push=False,
hub_model_id=None,
hub_private_repo=False,
hub_strategy=HubStrategy.EVERY_SAVE,
hub_token=<HUB_TOKEN>,
ignore_data_skip=False,
include_inputs_for_metrics=False,
jit_mode_eval=False,
label_names=None,
label_smoothing_factor=0.0,
learning_rate=0.0001,
length_column_name=length,
load_best_model_at_end=False,
local_rank=0,
log_level=passive,
log_level_replica=warning,
log_on_each_node=True,
logging_dir=somefile,
logging_first_step=False,
logging_nan_inf_filter=True,
logging_steps=500,
logging_strategy=IntervalStrategy.STEPS,
lr_scheduler_type=SchedulerType.LINEAR,
max_grad_norm=1.0,
max_steps=-1,
metric_for_best_model=None,
mp_parameters=,
no_cuda=False,
num_train_epochs=10,
optim=OptimizerNames.ADAMW_TORCH,
optim_args=None,
output_dir=somedir,
overwrite_output_dir=False,
past_index=-1,
per_device_eval_batch_size=8,
per_device_train_batch_size=12,
prediction_loss_only=False,
push_to_hub=False,
push_to_hub_model_id=None,
push_to_hub_organization=None,
push_to_hub_token=<PUSH_TO_HUB_TOKEN>,
ray_scope=last,
remove_unused_columns=True,
report_to=[],
resume_from_checkpoint=None,
run_name=somedir,
save_on_each_node=False,
save_safetensors=False,
save_steps=500,
save_strategy=IntervalStrategy.STEPS,
save_total_limit=None,
seed=7102,
sharded_ddp=[],
skip_memory_metrics=True,
tf32=None,
torch_compile=False,
torch_compile_backend=None,
torch_compile_mode=None,
torchdynamo=None,
tpu_metrics_debug=False,
tpu_num_cores=None,
use_cpu=False,
use_ipex=False,
use_legacy_prediction_loop=False,
use_mps_device=False,
warmup_ratio=0.0,
warmup_steps=1,
weight_decay=0.0,
)
2

This post is probably affected by this issue reported on Github: you may want to ping the HF maintainers there to let them know, if you are affected by this issue.