Hello,
Based on conversations information found on this channel (thanks to @eugr @johnny_nv) and the internet, I’ve consolidated a playbook to run Nemotron Nano VL 12B V2 at NVFP4 quantization on your DGX spark using vLLM:
Feel free to test it and contribute. I’ll add more later this week.
Thank you for sharing @raphael.amorim !
I have moved this topic to DGX Spark / GB10 User Forum > DGX Spark / GB10 Projects
noice!
It also seems to work for nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-NVFP4-QAD that came out today https://arxiv.org/pdf/2511.03929
ETA:
it looks like nvidia/Nemotron-Nano-VL-12B-V2-FP4-QAD redirects to nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-NVFP4-QAD – why use the former vs the latter? Also, hugging face cache does not seem to know they are the same, so now I have two copies under different names.
models--nvidia--Nemotron-Nano-VL-12B-V2-FP4-QAD
models--nvidia--NVIDIA-Nemotron-Nano-12B-v2-VL-NVFP4-QAD
I’ve actually optimized load time by 7x by using fastsafetensors library in the latest commit
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.