Decode · tok/s
103
Prefill
2092
tok/100W
42.9
update! deepseek v4.1 flash just got faster at reading v1.1 of my 2x dgx spark recipe is out and the prompt side is up to 18% quicker 🐋 > 8k prompt, 1,763 tok/s → 2,090 tok/s prefill now > 32k prompt, 1,773 tok/s → 2,092 tok/s prefill now > decode unchanged, 103.0 / 57.9 / 141.0 tok/s on code / prose / structured, dspark on
First-person 2× DGX Spark. Rank is the quoted code 103.0. Prose 57.9 and structured 141.0 stay notes. Prefill rank is the new 2,092 at 32k. Weights smaller than the prior 2.9 bpw pack.