Nimble
A 9B Qwen3.5 fine-tune for typed decisions, with tooling to curate contrastive training pairs that differ in exactly one fact.
0 members list Nimble, unchanged over the last 12 weeks, data as of 7 November 2026
Members who list Nimble in their Stack, by week.
Members who list Nimble
Be the first to add Nimble
Nobody on Stackness runs Nimble yet. Add it to your stack and you're the first member running it.
About Nimble
Nimble is a large language model based on Qwen, designed to make typed decisions in applications. It functions as a fine-tuned variant optimized for tasks that require structured outputs with specific data types. Developers use it alongside tooling for creating contrastive training datasets, where pairs of examples differ in exactly one factual detail, enabling more precise model training. The project is developed by bespokelabsai and is written in Python, offering local execution capabilities for decision-making tasks.
The tool includes functionality for dataset curation and model evaluation to support developers in improving performance on their specific decision problems. By working with contrastive training pairs, it helps surface how model behavior changes when individual facts shift, providing a systematic approach to understanding and refining model outputs for typed decision scenarios.
Checked 2 November 2026.
Nobody on Stackness runs Nimble yet.
Sign up and add itPosts about Nimble
- Strands Decider 2B: what it is and where it sits in an agent harness
Strands Decider 2B is AWS's free 1.9B decision model, out 1 October 2026: about 115 ms per question on an RTX 3090, consulted before each tool call.
- Open-source alternatives to Jev, and what each one replaces
The open-source alternatives to Jev are Laya, Kev, SemIf, Open-Jev and Nimble. They are four kinds of project, and here is what each one replaces.