Ternary Bonsai 2 27B NInfer-3090
Community native NInfer conversion of Prism ML's ternary Bonsai 2 27B, with vision and speculative-decoding components for the RTX 3090 fork.
What happened lately
Ternary Bonsai 2 27B NInfer-3090 timeline
This community conversion packages Prism ML’s existing Ternary Bonsai 2 27B text weights with a Qwen3.8 vision tower and ProCreations MTP and DFlash2 components. The October 15, 2026 upload provides an 8.87 GiB native .ninfer file, a conversion report and a manifest with source revisions and hashes. It is not a GGUF file or a Transformers checkpoint.
The model card instructs users to build the RTX 3090 fork of NInfer and download the bonsai2-27b artifact. Its performance figures are the author’s few-prompt measurements on one 24 GB RTX 3090 under Windows with CUDA 12.8. AITrending has not reproduced them. Local inference requires compatible hardware and runtime software; the runtime does not itself execute generated tool calls.