projects / picoclvr.git / blob
? search:


223565e63ad6236d33f90fa38e18d8b15e9d3193
[picoclvr.git] / README.txt
1 For the stack experiment:
2
3 ./main.py --task=stack --stack_fraction_values_for_train=0.75 --stack_nb_stacks=3
4
5 Takes ~1h10min on a 4090.
A series of toy tasks and a cache-equipped GPT-like model for fast autoregression.