logoalt Hacker News

mathisfun123today at 5:50 PM1 replyview on HN

lol the same way we implement all of the reduced precision fp8, fp4 types today: by storing them in the corresponding uint:

https://github.com/ggml-org/llama.cpp/discussions/15095


Replies

noosphrtoday at 6:03 PM

Balanced ternary fp is not a reduced precision type of binary fp: https://arxiv.org/abs/2512.10964

>Unlike their binary counterparts, posits and takums, tekums simultaneously accommodate both ∞ and NaR, while retaining the simplicity of negation by flipping the underlying trit string. Perhaps most strikingly, tekums enable rounding by truncation, a property that eradicates at a stroke some notorious problems of rounding in binary arithmetic: double rounding errors, cascading carries in hardware, and the attendant inefficiencies.

show 1 reply