Gemma 4 26B A4B is the most served model on Darkbloom right now.
You serve compute on Darkbloom. Then you rally in MLX.fast to make compute faster.
Let's turn MLX.fast into the open R&D arm for Darkbloom.
LIVE on @yukonresearch:
this is exactly why graphjin.com is now a training environment it’s easy to SFT or RL a small model or a LORA with it on your companies own databases to get a specialized beast of a small model that can do millions of task for you cheap and fast
Training tiny models for special purpose use cases works so incredibly well if you have a great self improving recursive flywheel. Shopify ML team is on fire.
finetuned 0.8b model beats GPT 5.6-sol xhigh in this very specialized task.
