Trendoday

Precision on Demand: Escalating Weight Bits at Uncertain Decoding Steps in a Small Local Model

Local decoding usually fixes one weight precision for the whole generation. We study a different allocation: store 8-bit weights, decode with their top 4 bits, and redo a step at 8 bits only when the

  • Upvotes5
–Rank now
#25Peak rank
5Days on list
61Score · 7d
Nov 12, 2026First seen
Rank · 7 days

About this paper

A dynamic weight precision allocation strategy is proposed, utilizing low-bit quantization and uncertainty-based gating to selectively upgrade weight precision during LLM generation, thereby optimizing computational cost.

Local decoding usually fixes one weight precision for the whole generation. We study a different allocation: store 8-bit weights, decode with their top 4 bits, and redo a step at 8 bits only when the low-precision output distribution looks uncertain (entropy above a threshold or a small top-two margin). Quantization is simulated, the model is Qwen3-0.6B, and the benchmark is GSM8K. Cost is counted in weight bytes read per token. On 36724 replayed steps the entropy of the low pass ranks argmax flips with AUC 0.93. At a 15% escalation budget it catches 78% of flips and the margin gate 94%, against the 15% expected for a random gate of equal rate (a computed expectation, not a run). In 150…

weight precisionquantization8-bit weightstop 4 bitsentropymargin gateescalation budgetKL divergenceQwen3-0.6BGSM8K
Upvotes
5
Authors
3
Published
Nov 10, 2026

Summary and abstract excerpt via Hugging Face Papers ↗ · arXiv.

Trending history

Precision on Demand: Escalating Weight Bits at Uncertain Decoding Steps in a Small Local Model was first seen on Hugging Face · Daily papers on Nov 12, 2026. Since then it has appeared in 26 checks on 5 different days. Its best position was #25 of 30, reached on Nov 13, 23:28 UTC. It dropped off the list after Nov 16, 03:28 UTC.

Rank history

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

#1#11#20#30 Nov 12Nov 13Nov 14Nov 15Nov 16 Nov 12, 19:28 UTC · #29Nov 12, 21:28 UTC · #30Nov 12, 23:28 UTC · #30Nov 13, 01:28 UTC · #29Nov 13, 03:28 UTC · #30Nov 13, 09:28 UTC · #29Nov 13, 11:28 UTC · #27Nov 13, 13:28 UTC · #27Nov 13, 15:28 UTC · #27Nov 13, 17:28 UTC · #27Nov 13, 19:28 UTC · #27Nov 13, 21:28 UTC · #26Nov 13, 23:28 UTC · #25Nov 14, 01:28 UTC · #26Nov 14, 03:28 UTC · #26Nov 14, 05:28 UTC · #27Nov 14, 07:28 UTC · #28Nov 14, 09:28 UTC · #28Nov 14, 11:28 UTC · #28Nov 14, 13:28 UTC · #29Nov 14, 15:28 UTC · #29Nov 14, 17:28 UTC · #30Nov 15, 01:28 UTC · #26Nov 15, 03:28 UTC · #26Nov 15, 05:28 UTC · #26Nov 16, 03:28 UTC · #29

Last 14 days

Best rank per day (UTC).

Nov 3today

Metric history

Platform metric as reported by Hugging Face.

0235 Nov 12Nov 13Nov 14Nov 15Nov 16 Nov 12, 19:28 UTC · 4Nov 12, 21:28 UTC · 4Nov 12, 23:28 UTC · 4Nov 13, 01:28 UTC · 4Nov 13, 03:28 UTC · 4Nov 13, 09:28 UTC · 4Nov 13, 11:28 UTC · 4Nov 13, 13:28 UTC · 4Nov 13, 15:28 UTC · 4Nov 13, 17:28 UTC · 4Nov 13, 19:28 UTC · 4Nov 13, 21:28 UTC · 4Nov 13, 23:28 UTC · 4Nov 14, 01:28 UTC · 4Nov 14, 03:28 UTC · 4Nov 14, 05:28 UTC · 4Nov 14, 07:28 UTC · 4Nov 14, 09:28 UTC · 4Nov 14, 11:28 UTC · 4Nov 14, 13:28 UTC · 4Nov 14, 15:28 UTC · 4Nov 14, 17:28 UTC · 4Nov 15, 01:28 UTC · 5Nov 15, 03:28 UTC · 5Nov 15, 05:28 UTC · 5Nov 16, 03:28 UTC · 5

Questions

Is Precision on Demand: Escalating Weight Bits at Uncertain Decoding Steps in a Small Local Model trending on HF Daily Papers right now?

Not at the moment. It was last on the list on Nov 16, 03:28 UTC, after 5 days of trending.

What is the highest rank Precision on Demand: Escalating Weight Bits at Uncertain Decoding Steps in a Small Local Model reached?

#25 on Hugging Face · Daily papers, on Nov 13, 23:28 UTC. Trendoday checks this list every 120 minutes.

How long has Precision on Demand: Escalating Weight Bits at Uncertain Decoding Steps in a Small Local Model been trending?

It was first seen on Nov 12, 2026 and has been on the list on 5 days (26 checks) since.