Decode · tok/s
65
Prefill
—
tok/100W
27.1
code at 65 tok/s on 1 spark vs 58 tok/s on my 2. tensorfold ran a 4 bit qad build of qwen 3.8 flash next on one spark … switch them off and it sits at 24 tok/s
First-person overnight run. Rank is the quoted 65 code on one Spark. 24 is drafts off. 28.2 is the 128k decode in the follow-up, not the rank.
