Robert Flynn

Results 3 issues of Robert Flynn

Hey lucidrains, firstly thanks for providing this implementation! I've noticed in the ConformerConvModule the inner dimension is increased by an expansion_factor which defaults to 2 `inner_dim = dim * expansion_factor`...

### Describe the bug Had my run crash due to logging gradient data. Is there a reason why there are no try except clauses so wandb crashing never crashes training...

cli

Currently, from the way the library is written conv_state and ssm_state are only used when generating one step at a time using the InferenceParams. It would be useful to use...