She's had some further work on code and math, improving honesty in conversations about facts she knows (from mostly flipping on first-answer pressure), and reducing role confusion.
Warning: RNN, still limited post-training. Capabilities in basic/quick benchmarks lean towards somewhat older instruct models (Qwen 2.5 7B, Llama 3 8B)
- Downloads last month
- 73
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support