When I tested tps with the existing 1 spark
The speed was not bad for its size.
2 sparks
The speed was much better than I thought.
I measured it again without thinking, and it came out a little better.
If there is more optimization, I think it will be over 50.
I wanted to test glm5.3-flash, but the optimization recipe hasn't come out yet, so I'll pass for now.
I tried to test deepseek-v4-flash, but the agent kept wandering... Maybe because the vision-based model is not optimized yet.
\(ㅇㅁㅇ)/ I like Spark
