2026  5

November  1

Directional Ablation, and How Astra Fixed Qwen3.5-0.8B’s Reasoning Loops

November 4, 2026 · 18 min · liqq

October  4

Fixing Softmax and Cross-Entropy: Two Numerical Pitfalls

October 25, 2026 · 3 min · liqq

Rotation Representations Part 1: Rotation Matrices, RoPE, and 6D/9D

October 24, 2026 · 9 min · liqq

Why Training Loss Sawtooths Across Epochs

October 15, 2026 · 14 min · liqq

From 1000 Steps to 1: Building Diffusion and Flow Models from Scratch

October 9, 2026 · 37 min · liqq