Examples

Run the dataset demo:

PYTHONPATH=$(pwd) python examples/train_on_datasets.py \
  --copy_num 64 --copy_len 128 --copy_vocab 64 \
  --br_num 64 --br_len 128 --br_depth 32 --seed 0

Figures are saved under transformer_instant/examples/figures/ by default.

Benchmark specific tasks via the standalone script:

python -m transformer_instant.examples.benchmark --task copy --n 128 --L 256 --V 64 --seed 0
python -m transformer_instant.examples.benchmark --task reverse --n 128 --L 256 --V 64 --seed 0

Deep ELM sweep via CLI (saves plots under transformer_instant/examples/figures):

exactstep deep-elm-bench --depths 2,3,4 --hidden 128,256,512 --lambdas 1e-4,1e-3,1e-2