Vlado Boza

Results 22 comments of Vlado Boza

I would really love this feature. So instead of this: ``` unsafe { let mut error_p: *mut i8 = ptr::null_mut(); cpp!([..., mut error_p as "char*"] { try { foo(); }...

Just noticed that it uses wrong loss. This if hasattr(model, 'seqdist'): from https://github.com/nanoporetech/bonito/blob/07f885ee9a1c0fef66e8177f00615c12128f453d/bonito/cli/train.py#L71 Should be model.module for distributed setting, or decided before dataparallel thingy is in place.

Yep, 4-5 days seem doable. Also it needs to patch TNC to NTC dimensions (due to dataloaded behaviour). Never mind, I will fix it for myself :)

Oh, I should install that AMP :)

Great, can we have that env var setting in the documentation? Although package feature flag would be even better, env var is enough. Thanks.

Couple of comments from regular user of RNNs: a) it is very hard to write proper library for neural nets, since the structure you want really varies (do you want...

Can confirm, this also happens in Tensorboard. This is our simplified script for reproducibility: Sagemaker script: ``` from sagemaker.pytorch.estimator import PyTorch from sagemaker.debugger import TensorBoardOutputConfig TENSORBOARD_S3_PATH = "s3://something/test-tb/" def create_estimator():...

Dear maintainers. Given that your product is right now unusable for any longer training on spot instances (checkpointing is broken) and realtime monitoring is impossible without external services (since TB...

Edit: Bug is still present. But there might be a difference between running m5.large (bug not present) instance and g4dn.2xlarge instance (bug is present). Edit2: Bug is present in m5.large...

As of today, the bug is still present.