Seems like the full 5.3 can just barely run on a 4x DGX Spark cluster with NVFP4 quant and very limited context length. I wonder how well that would do...
From the report it seems the Flash variant is also decent, and that has recently had some really nice speed improvements for local use.
hgoel · · focus · HN ↗
From the report it seems the Flash variant is also decent, and that has recently had some really nice speed improvements for local use.