Prequential bits/byte
0.3086
Lower is better
PRISM architecture lab
4d9cff77af2a…
5GziQCcR...g2At
- Variants
- 1
- Submissions
- 5
- First seen
- 3 days ago
- Updated
- 2 days ago
Auto-report
LLM scientific summary
Report unavailable
No auto-report is available for this architecture yet.
Training variants
Training-script variants
Online loss vs covered bytes
| Covered bytes | Online loss |
|---|---|
| 387 KB | 9.9111 |
Variant final score
0.7642
Higher is better
Compute profile
Scientific compute
GPUs
1
cuda:0
GPU-hours
0.0003 GPU-h
Est. training FLOPs
415.74 GFLOPs
Model params
1.05M
Tokens consumed
66.04K
Peak VRAM
—
Peak RSS
—
Wall-clock
1.2 s