Related articles

production_transformer.py in 2026 — what the code actually is now

The 2024 article on production_transformer.py is correct as transformer theory but doesn’t describe the code as it stands in 2026. Three transformer files now live in the same repo (teaching minimal, single-file pre-norm v1, RAGE-flavored v1.1 with RMSNorm + SwiGLU + GQA + RoPE + KV cache), shipped via IPFS ModelPack with sha256 verification. Here is the operational ground truth.

Learn More

Gödel

core choice logging and self-improvement readiness Current state To show that mindX is or is not a Gödel machine, we need a single, accurate log of core choices: what was perceived, what options were considered, what was chosen, why, and (when available) outcome. 1. Gödel choice schema and global log 2. Instrument core decision points 3. Ollama-driven self-improvement readiness 4. API and UI (optional) 5. File and dependency summary Area File(s) Change Core directive docs/survive.md […]

Learn More

The 60-Second AOT Autotune Probe — How mindXtrain Pins MI300X Performance Before Training Starts

Day 2 of the AMD × lablab.ai Developer Hackathon. The 60-second AOT autotune probe — the layer that mindXtrain is built around — runs on real MI300X silicon for the first time. This post explains what the probe measures, why “AOT-only” is the discipline that matters, and how the probe’s output flows into the rest of the pipeline so that training is reproducible across machines and across runs. 1. What the probe is, and what […]

Learn More