mirror of
https://github.com/allaunthefox/Research-Stack.git
synced 2026-07-31 03:05:21 +00:00
c54a0199be
7 concrete changes to enwik compression: - Shear pre-transform: 15-30% on structured regions (40% of enwik) - S3C shell position encoding: 5-10% positional overhead reduction - OAC speculative motifs: avoids 2-5% bloat, enables aggressive testing - FAMM preshaped context: 10-20% context efficiency gain - PIST n-D token encoding: 5-8% with cross-position probability sharing - Gram matrix dictionary: MB → KB overhead - Metric entropy coding: 10-15% entropy reduction in structured regions The Big Fold: 4 separate Hutter components collapse into 1 shear matrix. Estimated 12-22% overall compressed size reduction. |
||
|---|---|---|
| .. | ||
| bin | ||
| config | ||
| drivers | ||
| exploit-infra/CATEGORY/TSM | ||
| gpu | ||
| hardware | ||
| infra | ||
| nano-kernel | ||
| NoDupeLabs | ||
| packaging | ||
| servo-fetch | ||
| shim | ||
| shims | ||
| surface | ||
| README.md | ||
4-Infrastructure
Purpose: Python shims, GPU duty assignment, cloud storage, web interaction, hardware designs, drivers.
Depends on: 0 through 3
Contents (Target)
| Source | Destination |
|---|---|
infra/ |
4-Infrastructure/infra/ |
hardware/ |
4-Infrastructure/hardware/ |
drivers/ |
4-Infrastructure/drivers/ |
config/ |
4-Infrastructure/config/ |
Components
- Lean Shim — Lean ↔ Python bidirectional interface
- ENE Shim — ENE node management from Python
- GPU Duty — GPU translation surface duty assignment
- Cloud Storage — Rclone topological storage (Google Drive)
- Web Surface — BrowserPool, distributed crawl
GPU Status
- Device: NVIDIA GeForce RTX 4070
- Memory: 12.3GB total, ~10GB available
- CUDA: 13.0