PR #2061

open

Add 9L1216 int8 insurance non-record submission

by texanfirstView on GitHub
val_bpb
1.2635
Architecture
Transformer
Optimizer
Artifact Size
15,746,479 bytes

Training Techniques

Quantization
int8
bits: 8
scope: model weights

Novel Contributions

  • Non-record/fallback submission preserved as the safest reproducible artifact
  • Int8 compressed model artifact under the 16MB cap
  • Includes reproducibility materials such as training script, log, manifest, and checksums