Nokia Open-Sources AnyJev: Turn Any LLM into a Calibrated Decision Model, No Training
kalyan_kpl · x · 2026-09-25
Nokia Applied Research open-sourced AnyJev (578 GitHub stars), a framework that turns any LLM into a Jev-style decision model with typed decisions and real calibrated probabilities — no training required.
- Authored by Nokia (Sunnyvale) and Tencent Hunyuan researchers, with a paper.
- All levels serve via vLLM: at L2, an embed server's pooler hands back hidden states read by a closed-form head of just a few kilobytes, forming a decision endpoint.
- One command, python -m anyjev.pipeline <model>, converts, serves and measures in one go.
- Demonstrated with Qwen3-8B on a real BANKING77 item.
More from coding & agent
- Building a dynamic model router with LangChain and Jev middleware — Arindam_1729 · 2026-09-25
- Agentic RL inside the harness: how LFM2.5 trains with sandboxes and tools — helloiamleonie · 2026-09-25
- On-device agents need 3 things: ~4GB usable RAM, prefill speed, consistent quality — helloiamleonie · 2026-09-25
- Codex user reports quota bug: scheduled tasks burn 30% of $200 plan overnight — sethlazar · 2026-09-25
- The one-shot SaaS experiment seems to be working, reports indie dev gregmushen — gregmushen · 2026-09-25
- Claude Code users urged to run '/claude-api prompt-audit' to tune configs for Opus 5.5 — JeremyNguyenPhD · 2026-09-25