TrendBoard
Hugging Face · Daily papersoff the list

Specter: Fast MoE Inference via Quantization-Enabled Multi-Step Speculative Offloading

Mixture-of-experts (MoE) models often exceed a single accelerator's memory capacity. Offloading experts to host memory creates a severe PCIe I/O bottleneck because data-dependent routing keeps transfe

1 comments

Open on Hugging Face ↗Hugging Face Daily Papers: trending AI research
–Rank now
#10Best rank
–Score · 24h
206Score · 7d
2Days on list
10Snapshots
Oct 21, 2026First seen

Summary

Specter: Fast MoE Inference via Quantization-Enabled Multi-Step Speculative Offloading was first seen on Hugging Face · Daily papers on Oct 21, 2026. Since then it appeared in 10 hourly checks across 2 days, peaking at #10 on Oct 22, 01:18 UTC. It was last on the list on Oct 22, 01:18 UTC.

Rank history

Position on Hugging Face · Daily papers at every check (lower is better).

#1#11#20#30 07:1811:4816:1820:4801:18 Oct 21, 07:18 UTC · #15Oct 21, 09:18 UTC · #16Oct 21, 11:18 UTC · #17Oct 21, 13:18 UTC · #14Oct 21, 15:18 UTC · #13Oct 21, 17:18 UTC · #11Oct 21, 19:18 UTC · #12Oct 21, 21:18 UTC · #12Oct 21, 23:18 UTC · #11Oct 22, 01:18 UTC · #10

Last 14 days

Best rank per day (UTC).

Oct 14today

Metric history

Platform metric as reported by Hugging Face.

061319 07:1811:4816:1820:4801:18 Oct 21, 07:18 UTC · 11Oct 21, 09:18 UTC · 15Oct 21, 11:18 UTC · 15Oct 21, 13:18 UTC · 16Oct 21, 15:18 UTC · 17Oct 21, 17:18 UTC · 19Oct 21, 19:18 UTC · 19Oct 21, 21:18 UTC · 19Oct 21, 23:18 UTC · 19Oct 22, 01:18 UTC · 19

Embed badge

Show that this trended, linking back to this page.

Specter: Fast MoE Inference via Quantization-Enabled Multi-Step Speculative Offloading on TrendBoard 
<a href="https://trendoday.com/hugging-face/papers/2511.14102-specter-fast-moe-inference-via-quantization-enabled-multi"><img src="https://trendoday.com/badge/hugging-face/papers/2511.14102-specter-fast-moe-inference-via-quantization-enabled-multi.svg" alt="Specter: Fast MoE Inference via Quantization-Enabled Multi-Step Speculative Offloading on TrendBoard" height="44"></a>
[![Specter: Fast MoE Inference via Quantization-Enabled Multi-Step Speculative Offloading on TrendBoard](https://trendoday.com/badge/hugging-face/papers/2511.14102-specter-fast-moe-inference-via-quantization-enabled-multi.svg)](https://trendoday.com/hugging-face/papers/2511.14102-specter-fast-moe-inference-via-quantization-enabled-multi)