Lawyer with too many GPUs, part 4: DeepSeek-V4.1-Flash on eight V100s from 2018. 200+ tok/s on code, ~110 on legal writing, 2,200 tok/s prefill at 128K, 524K context, every weight in VRAM
fuck you sam altman. i am keeping my spare capacity. Update from the lawyer with the V100 server. Three days ago I got excited and posted DeepSeek-V4.1-Flash on four V100s at 36-38 tok/s, before I had done my custom kernel work and prefill optimizations. I also said I'd try it across eight. I did.