Hugging Face's logo
Activity Feed Request to join this org

AI & ML interests

None defined yet.
PeetPedroย 
posted an update 1 day ago
view post
Post
42
# ๐ŸŒŒ PeetPedro โ€” Hugging Face Sovereign AI Profile

**Identity:** Peter Lodri (PeetPedro / pocok0xRE / 0xp3t3rl)
**Focus:** BitNet b1.58 Ternary Weight Quantization, Metal GPU SIMD Operators, Sovereign AI Systems

---

## ๐Ÿš€ Spaces & Models
- **[PeetPedro/axiom-quant-demo]( PeetPedro/axiom-quant-demo: Live Gradio interactive space demonstrating BitNet b1.58 ternary weight operator algebra and SIMD contractions.

---

## ๐Ÿ“Š Datasets & Research Papers
- **[PeetPedro/axiom-quant-constellation]( PeetPedro/axiom-quant-constellation: Research monograph series PDF, TeX sources, and 7 high-resolution architectural infographics.
- **[PeetPedro/ultrawhale-dogfood]( PeetPedro/ultrawhale-dogfood: 24.8 MB synthetic Q&A fine-tuning dataset for BitNet b1.58 dogfooding.
- **[PeetPedro/kompress-ultra-bitnet-benchmarks]( PeetPedro/kompress-ultra-bitnet-benchmarks: Layer-by-layer weight sparsity & SVD energy decay JSON metrics.

---

## ๐ŸŒ External Constellation Links
- **Public Innovation Portal:** https://portail-vaked-dev.pages.dev/showcase.html
- **WASM SIMD Sandbox:** https://portail-vaked-dev.pages.dev/sgc_wasm_demo.html
- **Research Monographs:** https://axiomquant.org/
- **GitHub:** https://github.com/peterlodri-sec

*the constellation ยท 0 + 1 ยท fine touch from within ยท vaked.dev*
  • 4 replies
ยท
PeetPedroย 
posted an update 18 days ago
PeetPedroย 
posted an update about 1 month ago
PeetPedroย 
posted an update about 2 months ago
view post
Post
163
hey, I'm doing some experimenting, looping around :slight_smile:
---
**kompress-v6** *shipped* โ€” trained on Claude Code agent patterns (bash output, file reads, stack traces, search results, JSON tool responses). 3k synthetic pairs + 2k existing, fine-tuned from v4, $0.20 on vast.ai.

Results:
heretic exact_pct 0.962 (v4: 0.967),
keep_rate 0.854 (v4: 0.823),
override delta 0.
Model got more conservative โ€” higher keep_rate on structured technical content.
Real proxy:
v4 compressed 9.5%,
v6 compressed 4.2% on the same session.
Less aggressive, fewer must-keep tokens dropped on paths and identifiers.

Interesting failure: self-labeling with v4+override collapsed mk_in_ref to 0.652.
TokenExpiredError splits into Token+Expired+Error โ€” subtokens that don't individually match the must-keep regex, so the force-keep never fires. Generator references (mk_in_ref=1.0 by construction) ended up being better labels than v4's compressed output for agent data.
Fix for next run: slide a 2-3 subtoken window instead of checking individual subtokens. Would let self-labeling work on agent content and potentially produce a more compression-aggressive v7.

Models on HF:
- PeetPedro/kompress-v6
- PeetPedro/kompress-v4
- PeetPedro/kompress-v3
Write-up: https://pocoo.vaked.dev/posts/2026-06-25-kompress-v6-agent-distribution