Run Ornith-1.5-35B-A3B on your DGX Spark ✨
MoE shows its power on the spark, and it's really fast.
Up to 440 tok/s on 24 concurrent sessions
- b12x backend
- 256k context
- 24 concurrent sessions
- mtp
This should be better than Qwen3.6-35B, but not as good as Qwen3.8-27B. I'll be testing it out.
Thanks to @ViC305 for the base of the recipe.
Get it here:
https://t.co/OEuPAWSere