Robert Jan Schlimbach

[Question] Why hide dictconfig debugging content?

3

https://github.com/omry/omegaconf/blob/32e82d3d5ff2eebb7c55b1ba36ec084723f9f685/omegaconf/dictconfig.py#L320-L324 What is the reason for explicitly hiding information about the dictconfig while debugging?

[graph cmake] cmake utils in graph-dev vs master neither disjoint nor the same

1

# Summary By default, building oneDNN graph from source means also building oneDNN (original) from source, through `build_oneDNN.cmake`. Both `graph` and `original` contain a `cmake/utils.cmake`, which share a significant overlap...

[Feature request] Create PyPI build of ASAP

2

# Problem Working with the `multiresolutionimageinterface` python code is currently cumbersome, because no PyPI build of ASAP exists. This means ASAP doesn't integrate well with most python environment managers (pip/conda/venv/etc),...

[Feature] Improve error messages for IPEXConv

2

I was playing around with a debug model with some convolutions, but got the input shapes of some intermediate layer wrong (in my case a Conv after a Linear, i.e....

[Bug] BFloat16 + master_param_non_fused_step results in `None` grad

3

Observations: - Using fused optim (e.g. Adam) + `ipex.optimize(..., level='01', dtype=torch.bfloat16)` **no error** - non fused optimizer (e.g. AdamW) + `ipex optimize(..., level='01', dtype=torch.float32)` + `AMP bf16` **no error** -...

replaced distutils LooseVersion with packaging.version.parse

5

Fixes #4072, and #3996 partially. This is my first PR for easybuild-framework, help is appreciated :)

### TL;DR: - `tweak.py` obtains a EasyConfig's version parameter from `fetch_parameters_from_easyconfig`, but this doesn't work. - Advice: replace with a call to `process_easyconfig` - `tweak.py` compares EasyConfig's version parameter with...

Trace page in tensorboard remains empty

3

The trace view is one of my most important tools for profiling pytorch programs, but for the last few days I cannot manage to get it to display anything: ...

Figure out where the NaNs in autoencoder training are coming from:

1

Source: `2021-11-01/12-14-29/0/lightning_logs/version_8317429 ` 

Problem: - pl.seed_everything only seeds numpy.random, the new numpy.default_rng() is not seeded. - This leads to un-seeded crops, which are not desirable from a reproducibility perspective Solution: - Use pytorch.random