Dev-4B: quick calibrated decisions on a document, reason only when the router flags risk

Suhaas Teja released Dev-4B, a 4B-parameter local model built on Qwen3-4B-Instruct-2507 plus roughly 133 MB of add-ons that answers typed document questions (choice, yes-no, score) with calibrated confidence and escalates to chain-of-thought only when a built-in router predicts the quick answer is wrong. The model ships with a Hugging Face Space demo, a Hugging Face model card, and an MLX 8-bit build, and is released under non-commercial CC BY-NC-SA weights with no Ollama or LM Studio support because of its LoRA adapter, decision head and router. Teja states the model is weaker on unseen task types than trained ones, is English-only, and was trained on documents of about 4,000 characters, and is seeking feedback on Mac performance and router escalation behavior.

I released Dev-4B — a small local model that answers typed questions about a document choice / yes-no / score with calibrated confidence, and only thinks step by step when a built-in router predicts the quick answer is likely wrong. Space: Dev-4B - a Hugging Face Space by suhaas-teja https://huggingface.co/spaces/suhaas-teja/Dev-4B-demo Model: suhaas-teja/Dev-4B · Hugging Face https://huggingface.co/suhaas-teja/Dev-4B MLX 8-bit build is on the same profile as Dev-4B-MLX-8bit search that name on Hugging Face . Stack: Qwen3-4B-Instruct-2507 + ~133 MB of add-ons LoRA adapter switched off while reading the document and on for the question, decision head, temperatures for calibration, tiny router . Base generation is unchanged when add-ons are off. Results on 7,100 frozen test questions: Limits to be clear about: weaker on unseen task types than on trained ones; reasoning is still a 4B model; English; docs ~4k chars in training; NC weights CC BY-NC-SA ; no Ollama/LM Studio LoRA + decision head + router . Question format follows System One; not affiliated with TypeSafe. Most open System One clones are decision-only. Dev-4B’s wedge is local calibrated decisions plus a router that escalates to CoT when the quick pass looks wrong — for Mac indie / offline routing experimenters. Would love feedback from people running local decision models — especially Mac numbers and cases where the router should/shouldn’t escalate.