NovaLiveSystem v4.1 (archived research checkpoint)
Archived. Superseded by nova-mind-v5. This repository stays public because it has a permanent DOI (10.57967/hf/7399).
What this is
A LoRA adapter for dphn/Dolphin3.0-Qwen2.5-3b, published in December 2025 as an early step in the Nova Mind line of research models. It is a research checkpoint, not a product.
Scores withdrawn (October 2026)
An earlier version of this card reported results from an internal 22-question test set, including a "medical safety" result and a "production-ready" rating. The set was small and internal, and the results were not independently checked, so they do not support those claims. They have been withdrawn.
For results on standard public benchmarks, measured with EleutherAI's lm-evaluation-harness and checked for training-data overlap, see the successor model, nova-mind-v5.
Not for
Medical, legal, financial or other high-stakes use. This model has not been evaluated for safety.
The previous card remains in this repository's history.
- Downloads last month
- 31