PG Field Guide
Learn
Techniques
Emerging
PRs
← PR #1864
PR #1866→
PR #1865
open
val_bpb 0.87480 (3-seed mean) Donnie
by newjordan
View on GitHub
val_bpb
0.8748
Architecture
Transformer
Optimizer
—
Artifact Size
—
Training Techniques
Novel Contributions
3-seed mean submission
Reported validation bits-per-byte across seeds 42, 300, and 444