Blog

Notes and project write-ups.

Spectral Compact Training (Reimplementation)

A reimplementation study of Björn Kohlberger's Spectral Compact Training (2026) — every weight stored as a truncated SVD, shrinking LLaMA-70B per-MLP-layer training footprint 199× and fitting a full step inside 8 GB.

19 min read · May 11, 2026

2026 · machine-learning llm optimization reimplementation · projects
BRIDGE — Latent-Diffusion Skill Policies for Offline Multi-Task Manipulation

A reproducible offline-RL pipeline that lifts D4RL Franka-Kitchen task completion from 58.3% → 75.0% (+16.7 pts over BCQ) via a state-decoder-anchored β-VAE skill latent feeding an XQL-ranked latent diffusion policy.

3 min read · May 11, 2026

2026 · reinforcement-learning diffusion-models offline-rl pytorch · projects
Linear Regression: From Equation to Code

A short tour through the closed-form and gradient-descent solutions for linear regression — math, LaTeX source, and runnable NumPy.

6 min read · May 10, 2026

2026 · machine-learning python latex · notes
Sample Post: Markdown, Code, Math, and Images

A reference post showing how this blog renders developer-friendly content. Copy, edit, write your own.

5 min read · May 10, 2026

2026 · meta sample · notes