A DGX Spark-like, the Asus GX10. I’m kicking myself that I didn’t buy a second one when I thought about it months ago, but Nvidia’s NVFP4 quantisation of Flash Next and offloading the ngram table to NVMe has worked well
I have a half finished repo that sets it all up for production (using systemd, not hacky scripts and one-off docker commands), if someone is interested in this, tell me, and it might give me the push to actually finish up the latest threads and publish it!