src/dnadiffusion rewrite

[ssenan]

This is a large change to remove a loss spike issue that occurred after splitting single script version code into separate files

Resolves #149

Code changes

• Moves code back towards using builtin pytorch (with the exception of huggingface accelerate being used to assist with distributed training)
• Hydra-zen train loop is still a wip, so train_hf.py contains the main train call that can be linked to the slurm script for distributed training
• sample.py is used to load a checkpoint and generate cell-specific sequences for validation
• dnadiffusion.py contains all code in a single script that can also be used for training
• in the top directory of notebooks there are now two notebooks: master_dataset.ipynb and filter_master.ipynb, which show how our original data was collated for our complete table and then filtered down to our current working set
• Final major change is that diffusion functions have been collected into a class and this class has been integrated into the main trainloop (src/dnadiffusion/utils/train_util.py)
• There are a multitude of other small changes made to accommodate these larger changes

Code should now be more readable in the single script version dnadiffusion.py and more extensible in src/dnaddifusion

[codecov]

Codecov Report

Merging #150 (2203dc3) into main (b4ba5aa) will increase coverage by 1.04%.
The diff coverage is 0.00%.

@@           Coverage Diff            @@
##            main    #150      +/-   ##
========================================
+ Coverage   1.72%   2.76%   +1.04%     
========================================
  Files         18      12       -6     
  Lines       1278     795     -483     
  Branches     117      88      -29     
========================================
  Hits          22      22              
+ Misses      1256     773     -483     

Impacted Files	Coverage Δ
src/dnadiffusion/data/dataloader.py	0.00% <0.00%> (ø)
src/dnadiffusion/metrics/metrics.py	0.00% <0.00%> (ø)
src/dnadiffusion/models/diffusion.py	0.00% <0.00%> (ø)
src/dnadiffusion/models/layers.py	0.00% <ø> (ø)
src/dnadiffusion/models/unet.py	0.00% <0.00%> (ø)
src/dnadiffusion/utils/sample_util.py	0.00% <0.00%> (ø)
src/dnadiffusion/utils/train_util.py	0.00% <0.00%> (ø)
src/dnadiffusion/utils/utils.py	0.00% <0.00%> (ø)