LoRA Co-inventor Joins Mercor, Releases 397B RL Training Guide
himanshustwts · x · 2026-09-02
Edward Hu, co-inventor of LoRA and muP (ex-OpenAI), has joined Mercor to lead model training and research. Mercor released a detailed guide on post-training Qwen3.5-397B-A17B using reinforcement learning (RL) with SkyRL. Focusing on expert data for knowledge work agents, the project improved Pass@1 on the APEX-Agents benchmark from 16.11% to 27.29% (a 70% relative increase). The full training script, model weights, and evaluation traces have been open-sourced.
More from coding & agent
- LLM one-shots a Minecraft 3x3 piston door by building its own Linux test infra — Vjeux · 2026-10-18
- Dev builds Mumbai-inspired 3D manga background generator from OpenStreetMap data — CurieuxExplorer · 2026-10-18
- Solo founder builds a money-printing data service on Claude's Pi Durable harness — MicahBerkley · 2026-10-18
- Dev builds Arch-based OS with custom desktop and browser using Claude Code — Snoo-10493 · 2026-10-18
- Five SaaS agent workflows fail the same way — and it's never model intelligence — usehive · 2026-10-18
- Claude skill generates hand-drawn interactive diagrams that teach kids how things work — CatAstro_Piyush · 2026-10-18