tools
Arcee-SuperNova
Arcee-SuperNova is an open-weight 70B language model for instruction following, mathematical queries, and local or API deployment.
Arcee-SuperNova is a 70B language model based on Llama-3.1-70B-Instruct. Arcee says it combines distillation from Llama-3.1-405B-Instruct, synthetic instruction tuning, direct preference optimization, and model merging.
It is intended for developers, researchers, and organizations that need instruction-following or mathematical capabilities with control over deployment. The model weights are available under the Llama-3 license; hosting, inference infrastructure, and any API usage may cost extra. Arcee's current pricing page does not list SuperNova.
Features
- 70B-parameter language model
- Instruction following
- Mathematical query handling
- Distillation from Llama-3.1-405B-Instruct
- Synthetic instruction tuning with Evol-Kit
- Direct Preference Optimization (DPO)
- Model merging
- Llama-3 license for commercial and non-commercial use
Use cases
- Generate responses for general instruction-following tasks
- Solve mathematical queries
- Run the model on local infrastructure
- Serve the model through an OpenAI-compatible API
- Retrain the model with reinforcement learning from human feedback