Nace AI publishes 50-page playbook for training SOTA decision models, Drex is open source
m4rkGS · reddit · 2026-11-06
Nace AI Research Lab released a 50-page technical report detailing how its open-source Drex decision models are built, trained, and evaluated — the first full community playbook for decision models.
- Models: 9B–35B family spanning Dense Causal Transformers, Diffusion LMs, and MoE, with bidirectional/sparse attention and tuned pointer decision heads
- Training: full-parameter SFT on 1.15M verified decision records, consistency training for option ordering, and RLAF with verified outcomes; ablations show most gains came from SFT, while RLAF mainly improved calibration
- Eval: Drex v1.5 hits 93.4% accuracy on a 32K–128K long-context benchmark; Drex v2.0 has 1.37% calibration ECE on 115K held-out decisions; <100ms latency at 128K tokens; option shuffling flips the top answer in only 4–6% of cases
- Deployment: safetensors and GGUF with llama.cpp/Ollama support; report covers quantization, caching, and reproducibility
Limitations include closed-book knowledge and missing same-backbone ablations. Useful for local agent routers, tool selection, or training your own decision models.
More from coding & agent
- Agent builds cable modem emulator that speaks DOCSIS in Rust and boots unmodified firmware — moyix · 2026-11-06
- ChatGPT's 1.2B weekly users are already doing agentic tasks in main chat — omooretweets · 2026-11-06
- bas-guard: a local offline security scanner giving Claude Code setups an "antivirus for agents" — victor_explore · 2026-11-06
- TMail Open-Sourced: Private Email for AI Agents on Your Tailscale Network — evilsocket · 2026-11-06
- wesbos migrates 15-year WordPress site to Emdash with agent skills, page speed jumps 52 to 75 — ritakozlov · 2026-11-06
- LLVM Dev Meeting 2026: Building durable open-source infrastructure in the agentic era — clattner_llvm · 2026-11-06