Steven Gonsalvez

Software Engineer

what the hell a 27B model in ~5.9 GB Ternary Bonsai 2 27B - 1.76-bit ternary weights - 262K context - text + image input - coding + reasoning + agentic tasks - runs locally - Apache 2.0 licensed P

Why CEREBRO kept it

Ternary Bonsai 27B: local, quantized, agentic-capable; rising open-source framework.

The text below is an automated extraction of the article at https://x.com/itsjdraven/status/2103678685362761962, stored verbatim in the public cerebro-vault repository. Copyright remains with the original publisher (x.com).

Ternary Bonsai 2 27B

- 1.76-bit ternary weights - 262K context - text + image input - coding + reasoning + agentic tasks - runs locally - Apache 2.0 licensed

P

> Ternary Bonsai 27B: local, quantized, agentic-capable; rising open-source framework.

what the hell a 27B model in ~5.9 GB

Ternary Bonsai 2 27B

- 1.76-bit ternary weights - 262K context - text + image input - coding + reasoning + agentic tasks - runs locally - Apache 2.0 licensed

PrismML keeps 98.2% of the benchmark performance

27B class AI in a tiny package https://t.co/PZw7nI7MBp

Backlinks

Appeared in 1 briefing

Related

Shares tags: cerebro/signal · repo/trending · vibe-coding

Also from x.com