(test) test ResNet backbone

(cleanup) removing timm from the codebase.

(debug) to torch operations in bfloat conversions.

Eval to infer.py

(optim) benchmark=True and higher eval batch size

Speeding up eval by reducing iters

Added precision training option

(optim) prefetch and speed up eval.

(debug) 6 chan instead of 7. (warn: conflicting git changes)

to mit_b2 as per paper

(debug) eval fixes

(config) reducing training iters and eval, checkpoint intervals

Training loop and around the codebase

Project skeleton generated