basta.one
Notes on AI and robotics — interpretability, inverse RL, and things I build.
Recent posts
Lightning McQwen3: Making Qwen3-0.6B think more efficiently #llm#reasoning#post-training How I trained Qwen3-0.6B to use 47% fewer tokens for under one point of accuracy, with thinking still on, mostly on one 12 GB GPU.
Why this blog exists #meta A first post — what this site is for, and a check that code blocks and equations render.