val_bpb
1.1230
Architecture
11L XSA4
Optimizer
—
Artifact Size
20,906,280 bytes
Training Techniques
Architecture
XSA4
Uses an XSA4 model variant with 11 layers.
parameters: {"layers":11}
Weight Averaging
EMA
parameters: null
Quantization
int6
bits: 6
scope: all
Evaluation
sliding window eval
parameters: {"stride":64}
Compression
zstd
level: null
Novel Contributions
- 11-layer XSA4 model run
- EMA-weighted checkpoint
- Int6 quantized submission
- Sliding-window exact evaluation with stride 64
- Non-record run that exceeds the 16MB submission limit