Eureka Machine writes a Metal megakernel fusing Flint's entire decode pass
cong_ml · x · 2026-10-22
Recursive's automated research system, Eureka Machine, now writes entire megakernels: using Metal, Apple's GPU language, it fuses Flint's whole decode forward pass into a single kernel on M4/M5 Macs — every layer, attention, low-bit projections, expert routing and output head.
The code is open-sourced on GitHub (recursive-oss/rec, megakernel.py, 1337 lines). The attached image shows the actual source blocks: per-layer loop, 4-bit QKV projections, attention output merging, sparse expert routing handoff, and final norm + logits.
Related event: Recursive's Eureka Machine Moves Toward Recursive Self-Improvement(2 posts)→
More from Infra
- Ollama Runs Open Models on NVIDIA Blackwell via Crusoe, Live in Two Weeks — ollama · 2026-10-22
- Claim: every 2nd/3rd-tier inference provider now has a recipe for 90%+ compute margins — zephyr_z9 · 2026-10-22
- NVIDIA keynote: silicon photonics CPO switches hit 102.4 Tb/s, Vera Rubin architecture unveiled — thealexbanks · 2026-10-22
- PyTorch Foundation Adds Nine Members, Nebius, Nscale and Shopify Join as Platinum — PyTorch · 2026-10-22
- GPU startup Flappy Airplanes raises $1B Series A co-led by Index and Lux — vitaliychiley · 2026-10-22
- Crusoe unveils Tinker API-compatible LLM post-training platform at PyTorch Conference — PyTorch · 2026-10-22