[BUG] Deepspeed and RLHF causes error

[pascal-pfeiffer]

🐛 Bug

Running the default RLHF experiment with Deepspeed will fail with the following error:

Traceback (most recent call last):
File ".../train_wave.py", line 106, in
run(cfg=cfg)
File ".../train.py", line 948, in run
val_loss, val_metric = train_function(
File ".../train.py", line 484, in run_train_rlhf
train_dataloader.sampler, "set_epoch"
AttributeError: 'DeepSpeedDataLoader' object has no attribute 'sampler'