Everything we have published, newest first.
One shared 8B base plus a 50MB LoRA adapter per customer gives you 1,000 specialists without 1,000 GPUs.
Sep 1, 2026
•
3 min read
Traditional ML can't read a demo request, and a 70B LLM is too slow and costly to fix it. A fine-tuned 8B specialist can.
Aug 30, 2026
