Linear-attention language models keep their context in a recurrent state of fixed size, where an inference system stores one copy of this state for every sequence it decodes at the same time. Storing
3 comments
PACE: Periodic and Channel-Aligned Encoding of Four-Bit Recurrent States was first seen on Hugging Face · Daily papers on Oct 22, 2026. Since then it appeared in 11 hourly checks across 2 days, peaking at #10 on Oct 23, 03:18 UTC. It was last on the list on Oct 23, 05:19 UTC.
Position on Hugging Face · Daily papers at every check (lower is better).
Best rank per day (UTC).
Platform metric as reported by Hugging Face.
Show that this trended, linking back to this page.
<a href="https://trendoday.com/hugging-face/papers/2610.29545-pace-periodic-and-channel-aligned-encoding-of-four-bit"><img src="https://trendoday.com/badge/hugging-face/papers/2610.29545-pace-periodic-and-channel-aligned-encoding-of-four-bit.svg" alt="PACE: Periodic and Channel-Aligned Encoding of Four-Bit Recurrent States on TrendBoard" height="44"></a>
[](https://trendoday.com/hugging-face/papers/2610.29545-pace-periodic-and-channel-aligned-encoding-of-four-bit)