Attention — Oct 27, 2026
What caught my attention on Oct 27 (1 item).
Building a game asset editor in 12GB VRAM
NoPlaceLikeLocalhost (Article:09:58) - Published 2026-10-26 18:00 - Discovered 2026-10-27 09:58
In this video, the creator tests whether the small local model Ternary Bonsai 2 27B can build a useful game asset editor while staying within 12GB of VRAM. The model is notable for its tiny footprint: the PQ2 quantization is about 7.2GB and claims to retain most of the intelligence of a much larger Qwen 3.8 27B model. To run it, the creator uses a custom llama.cpp fork because stock llama.cpp cann…