← Back to Architecture

Value Residual Learning (VRL)

Architecture
Used in
2 PRs
Best BPB
1.1175
Avg BPB
1.1204

Submissions

PR #569by gowtham0992
1.1175
PR #657by anthony-maio
1.1234

Hyperparameters Across PRs

pr_numberparameters
569{"learned_alphas":10,"sigmoid_init":0}
657{"layers":11,"initial_gate_bias":-1.5,"initial_mixing":"approx 18%"}