Nemotron‑3‑Super 64B Math Model in GGUF Format Available on Hugging Face
Use the 64B Nemotron‑3‑Super math model in GGUF format for agentic coding and math tasks on consumer GPUs.
Run the model locally on your GPU to test math and coding performance.
Summary
A new 64‑billion‑parameter Nemotron‑3‑Super model fine‑tuned for math tasks has been released on Hugging Face. The model is packaged in GGUF format, enabling efficient inference on consumer GPUs. It incorporates A12B Math REAP tuning, making it especially strong at mathematical reasoning and agentic coding. The author reports successful use on a dual‑TITAN RTX setup with a 500k‑token context window.
The model is hosted under the Max‑and‑Omnis namespace on Hugging Face and can be downloaded directly for local deployment. Its GGUF quantization allows low‑memory usage while maintaining performance. Users can test the model for coding and math challenges without needing a 12‑billion‑parameter GPU. The release demonstrates that large‑scale math‑focused LLMs can run on modest hardware when properly quantized.
Key changes
- 64B Nemotron‑3‑Super model fine‑tuned for math tasks
- Packaged in GGUF format for efficient inference
- Uses A12B Math REAP tuning for enhanced mathematical reasoning
- Available on Hugging Face under Max‑and‑Omnis
- Supports 500k‑token context window on dual‑TITAN RTX
- Suitable for agentic coding and math challenges
- Low‑memory quantization enables running on modest GPUs