Lawyer with too many GPUs runs DeepSeek-V4.1-Flash on 2x DGX Spark; publishes full stack, raw data, and benchmarks
Read the original at www.reddit.com→Update from the lawyer with the V100 server. I went and got two DGX Sparks, because of course I did. The mid-life crisis remains on schedule. Same deal as before: still a lawyer, still driving the whole thing through...
Original headline: "Lawyer with too many GPUs, part 3: DeepSeek-V4.1-Flash on 2x DGX Spark + an old 3090 box. 2,600+ tok/s prefill to 256K, 322 tok/s structured output, full stack and raw data published"