Open jaseleephd opened 4 years ago
[ ] Make sure we get similar speed with the baseline (0 refinement, mean of the prior), on both codebases (lanmt-ebm, lanmt)
lanmt-ebm
lanmt
[ ] Do proper comparison between delta refinement and energy
Ablated n_layers for predicting the direction / magnitude
n_layers
[x] 4 / 4
[x] 3 / 3
[x] 2 / 2
[x] 1 / 1
[x] 4 / 2
[ ] Make sure we get similar speed with the baseline (0 refinement, mean of the prior), on both codebases (
lanmt-ebm
,lanmt
)[ ] Do proper comparison between delta refinement and energy