Errors in retrain for bindiff

[yasong]

Hi,

Thanks for your excellent work. But I encountered an issue when running

./scripts/bindiff.sh

training environment

torch                              1.10.0+cu111
torchaudio                         0.10.0+cu111
torchvision                        0.11.0+cu111
GPU: RTX 2080ti
RAM: 11GB

1. no argument "use_pooling" in @register_criterion('poj_similarity')

@register_criterion('poj_similarity')
class PojSimilarityLoss(FairseqCriterion):

    def __init__(self, args, task):
        super().__init__(args, task)
        self.inst_padding_idx = task.instruction_dictionary.pad()
        self.state_padding_idx = task.state_dictionary.pad()
        self.task = task
        self.args = args

    def forward(self, model, sample, reduce=True, train=True):
        no_state = self.args.no_state
        no_pce = self.args.no_pce
        pooling = self.args.use_pooling
        output = model(**sample['net_input'], masked_tokens=None, features_only=True, moco_head=False,
            moco_head_only_proj=False, lm_head=False, classification_head_name=None,
            has_state=not no_state, has_pce=not no_pce, pooling_instruction=pooling)

after changing this to

#pooling = self.args.use_pooling

pooling = self.args.no_pooling

got another error: 2. multiple values for keyword "has_pce"

  File "/mnt/g/Projects/OSCAR/model/fairseq/models/irbert/model.py", line 92, in forward
    x, extra = self.decoder(src, features_only, return_all_hiddens, moco_head=moco_head, has_state=has_state,
TypeError: IRBertEncoder object got multiple values for keyword argument 'has_pce‘

after removing this keyword, got another error: 3. got multiple values for keyword argument 'pooling_instruction'

  File "/mnt/g/Projects/OSCAR/model/fairseq/models/irbert/model.py", line 92, in forward
    x, extra = self.decoder(src, features_only, return_all_hiddens, moco_head=moco_head, has_state=has_state,
TypeError: IRBertEncoder object got multiple values for keyword argument 'pooling_instruction'

after removing this, got another error too:

        add_(Tensor other, *, Number alpha) (Triggered internally at  ../torch/csrc/utils/python_arg_parser.cpp:1050.)
  exp_avg.mul_(beta1).add_(1 - beta1, grad)
Traceback (most recent call last):
  File "train.py", line 356, in <module>
    cli_main()
  File "train.py", line 321, in cli_main
    main(args)
  File "train.py", line 95, in main
    train(args, trainer, task, epoch_itr)
  File "train.py", line 139, in train
    log_output = trainer.train_step(samples)
  File "/OSCAR/model/fairseq/trainer.py", line 346, in train_step
    raise e
  File "/Projects/OSCAR/model/fairseq/trainer.py", line 309, in train_step
    loss, sample_size, logging_output = self.task.train_step(
  File "/OSCAR/model/fairseq/tasks/fairseq_task.py", line 248, in train_step
    optimizer.backward(loss)
  File "/OSCAR/model/fairseq/optim/fp16_optimizer.py", line 103, in backward
    loss.backward()
  File "/lib/python3.8/site-packages/torch/_tensor.py", line 307, in backward
    torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs)
  File "/lib/python3.8/site-packages/torch/autograd/__init__.py", line 154, in backward
    Variable._execution_engine.run_backward(
RuntimeError: CUDA error: unknown error
CUDA kernel errors might be asynchronously reported at some other API call,so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.

I don't know how to fix this. Can you tell me what did I miss? Thanks for your time in advance.