SSLOnlineEvaluator does not work with DDP

[shubhamkulkarni01]

🐛 Bug

In commit 6e14209185c2b2100f3e515ee6782597673bb921 on pytorch_lightning from Feb 17, the use_ddp property was removed from AcceleratorConnector.

In commit b29b07e9788311326bca4779d70e89eb36bfc57f on pytorch_lightning from Feb 27, the use_dp property was removed from AcceleratorConnector.

The SSLOnlineEvaluator now throws exceptions with multiple GPUs since it checks for these properties in distributed training.

To Reproduce

Steps to reproduce the behavior:

Must run on a system with 2+ GPUs attached and accessible to PyTorch.

1. Create a pl.Trainer
2. Attach an SSLOnlineEvaluator Callback
3. Call trainer.fit

Code sample:

import torch
import pytorch_lightning as pl
import pl_bolts


def main():
    zdim = 2048
    bs = 8

    ds = pl_bolts.datasets.DummyDataset(
            (3, 224, 224),
            (1, ),
            num_samples = 100
    )
    dl = torch.utils.data.DataLoader(ds, batch_size=bs)

    model = pl_bolts.models.self_supervised.SimCLR(
            gpus = torch.cuda.device_count(),
            num_samples = len(ds),
            batch_size = bs,
            dataset = 'custom',
            hidden_mlp = zdim,
    )

# fit
    trainer = pl.Trainer(
        accelerator = 'gpu',
        devices = -1,
        callbacks = [
            pl_bolts.callbacks.SSLOnlineEvaluator(
                z_dim = zdim,
                num_classes = 4, # or any other number
                hidden_dim = None,
                dataset = 'custom'
            ),
        ],
    )

    trainer.fit(model, train_dataloaders = dl)
if __name__ == '__main__':
    main()

Leads to the following

Traceback (most recent call last):
  File "example.py", line 41, in <module>
    main()
  File "example.py", line 39, in main
    trainer.fit(model, train_dataloaders = dl)
  File "/opt/conda/lib/python3.7/site-packages/pytorch_lightning/trainer/trainer.py", line 604, in fit
    self, self._fit_impl, model, train_dataloaders, val_dataloaders, datamodule, ckpt_path
  File "/opt/conda/lib/python3.7/site-packages/pytorch_lightning/trainer/call.py", line 36, in _call_and_handle_interrupt
    return trainer.strategy.launcher.launch(trainer_fn, *args, trainer=trainer, **kwargs)
  File "/opt/conda/lib/python3.7/site-packages/pytorch_lightning/strategies/launchers/multiprocessing.py", line 117, in launch
    start_method=self._start_method,
  File "/opt/conda/lib/python3.7/site-packages/torch/multiprocessing/spawn.py", line 198, in start_processes
    while not context.join():
  File "/opt/conda/lib/python3.7/site-packages/torch/multiprocessing/spawn.py", line 160, in join
    raise ProcessRaisedException(msg, error_index, failed_process.pid)
torch.multiprocessing.spawn.ProcessRaisedException:

-- Process 1 terminated with the following error:
Traceback (most recent call last):
  File "/opt/conda/lib/python3.7/site-packages/torch/multiprocessing/spawn.py", line 69, in _wrap
    fn(i, *args)
  File "/opt/conda/lib/python3.7/site-packages/pytorch_lightning/strategies/launchers/multiprocessing.py", line 139, in _wrapping_function
    results = function(*args, **kwargs)
  File "/opt/conda/lib/python3.7/site-packages/pytorch_lightning/trainer/trainer.py", line 645, in _fit_impl
    self._run(model, ckpt_path=self.ckpt_path)
  File "/opt/conda/lib/python3.7/site-packages/pytorch_lightning/trainer/trainer.py", line 1083, in _run
    self._call_callback_hooks("on_fit_start")
  File "/opt/conda/lib/python3.7/site-packages/pytorch_lightning/trainer/trainer.py", line 1380, in _call_callback_hooks
    fn(self, self.lightning_module, *args, **kwargs)
  File "/opt/conda/lib/python3.7/site-packages/pl_bolts/callbacks/ssl_online.py", line 87, in on_fit_start
    if accel.use_ddp:
AttributeError: 'AcceleratorConnector' object has no attribute 'use_ddp'

Expected behavior

Environment

• PyTorch Version (e.g., 1.0): '1.13.0+cu117'
• Lightning version: '1.8.4.post0'
• pl_bolts version: '0.6.0.post1'
• OS (e.g., Linux): Docker (Ubuntu base)
• How you installed PyTorch (conda, pip, source): Pytorch Docker image
• Python version: 3.7.11
• CUDA/cuDNN version: 11.7
• GPU models and configuration: 2 A10s, 24GB VRAM each

Additional context

Currently have it patched in personal system as follows using the old definition of the use_ddp property prior to removal:

    from pytorch_lightning.trainer.connectors.accelerator_connector import _LITERAL_WARN, AcceleratorConnector
    AcceleratorConnector.use_ddp = lambda self: self._strategy_type in (
            _StrategyType.BAGUA,
            _StrategyType.DDP,
            _StrategyType.DDP_SPAWN,
            _StrategyType.DDP_SHARDED,
            _StrategyType.DDP_SHARDED_SPAWN,
            _StrategyType.DDP_FULLY_SHARDED,
            _StrategyType.DEEPSPEED,
            _StrategyType.TPU_SPAWN,
        )

[Borda]

@shubhamkulkarni01, what versions of PL and Bolts are you using?

[shubhamkulkarni01]

Lightning version: '1.8.4.post0'
pl_bolts version: '0.6.0.post1'

[Borda]

could you pls try the latest (today) 0.7.0 where we did some compatibility fixes