Block a user
research: quant strategy, and whether NVFP4 on the Blackwell cards buys real throughput
NVFP4 on our Blackwell: measured, on our exact card, by someone else
FlashML-org/FreeToken ships three NVFP4 expert-GEMM backends and picks
between them by capability and by expert width.…
A single file over 50GB cannot be fetched through rustingface by an unmodified client
A request that joins an in-flight fetch waits in silence for the whole transfer