Trendoday

SteerDistill: Quantizing LLM Agents via Closed-Loop Teacher Steering

Low-bit quantization reduces the inference cost of LLM agents but can compromise task success. Examining failed rollouts, we find that quantized agents overlook task requirements or environmental cond

  • Upvotes37
–Rank now
#4Peak rank
5Days on list
597Score · 7d
Nov 12, 2026First seen
Rank · 7 days

About this paper

Low-bit quantized LLM agents can recover much of their task success through closed-loop guidance from a full-precision teacher, followed by unassisted student training on corrected and reference turns.

Low-bit quantization reduces the inference cost of LLM agents but can compromise task success. Examining failed rollouts, we find that quantized agents overlook task requirements or environmental conditions, producing environment transitions that diverge from successful full-precision references. Correcting these behavioral divergences remains difficult: offline data may miss the student's failure contexts, while its own rollouts may provide few correct responses in those contexts. We introduce SteerDistill, a quantization-aware distillation method in which a full-precision teacher guides the student to revise failed responses and iteratively refines its guidance using feedback from…

low-bit quantizationLLM agentsquantization-aware distillationfull-precision teacherstudent modelrolloutsenvironment transitionsfull-precision referencesclosed-loop teacher steeringbehavioral corrections
Upvotes
37
Authors
6
Published
Nov 10, 2026

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

Trending history

SteerDistill: Quantizing LLM Agents via Closed-Loop Teacher Steering was first seen on Hugging Face · Daily papers on Nov 12, 2026. Since then it has appeared in 46 checks on 5 different days. Its best position was #4 of 30, reached on Nov 13, 05:28 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, 09:28 UTC · #12Nov 12, 11:28 UTC · #10Nov 12, 13:28 UTC · #9Nov 12, 15:28 UTC · #9Nov 12, 17:28 UTC · #7Nov 12, 19:28 UTC · #5Nov 12, 21:28 UTC · #5Nov 12, 23:28 UTC · #5Nov 13, 01:28 UTC · #5Nov 13, 03:28 UTC · #5Nov 13, 05:28 UTC · #4Nov 13, 07:28 UTC · #4Nov 13, 09:28 UTC · #6Nov 13, 11:28 UTC · #6Nov 13, 13:28 UTC · #6Nov 13, 15:28 UTC · #6Nov 13, 17:28 UTC · #7Nov 13, 19:28 UTC · #7Nov 13, 21:28 UTC · #7Nov 13, 23:28 UTC · #7Nov 14, 01:28 UTC · #7Nov 14, 03:28 UTC · #7Nov 14, 05:28 UTC · #7Nov 14, 07:28 UTC · #7Nov 14, 09:28 UTC · #7Nov 14, 11:28 UTC · #7Nov 14, 13:28 UTC · #7Nov 14, 15:28 UTC · #7Nov 14, 17:28 UTC · #7Nov 14, 19:28 UTC · #7Nov 14, 21:28 UTC · #7Nov 14, 23:28 UTC · #7Nov 15, 01:28 UTC · #7Nov 15, 03:28 UTC · #7Nov 15, 05:28 UTC · #7Nov 15, 07:28 UTC · #7Nov 15, 09:28 UTC · #7Nov 15, 11:28 UTC · #7Nov 15, 13:28 UTC · #7Nov 15, 15:28 UTC · #7Nov 15, 17:28 UTC · #7Nov 15, 19:28 UTC · #7Nov 15, 21:28 UTC · #7Nov 15, 23:28 UTC · #7Nov 16, 01:28 UTC · #7Nov 16, 03:28 UTC · #6

Last 14 days

Best rank per day (UTC).

Nov 3today

Metric history

Platform metric as reported by Hugging Face.

0132639 Nov 12Nov 13Nov 14Nov 15Nov 16 Nov 12, 09:28 UTC · 18Nov 12, 11:28 UTC · 22Nov 12, 13:28 UTC · 23Nov 12, 15:28 UTC · 23Nov 12, 17:28 UTC · 26Nov 12, 19:28 UTC · 30Nov 12, 21:28 UTC · 31Nov 12, 23:28 UTC · 31Nov 13, 01:28 UTC · 31Nov 13, 03:28 UTC · 32Nov 13, 05:28 UTC · 33Nov 13, 07:28 UTC · 34Nov 13, 09:28 UTC · 34Nov 13, 11:28 UTC · 34Nov 13, 13:28 UTC · 34Nov 13, 15:28 UTC · 35Nov 13, 17:28 UTC · 34Nov 13, 19:28 UTC · 35Nov 13, 21:28 UTC · 35Nov 13, 23:28 UTC · 35Nov 14, 01:28 UTC · 35Nov 14, 03:28 UTC · 35Nov 14, 05:28 UTC · 35Nov 14, 07:28 UTC · 35Nov 14, 09:28 UTC · 35Nov 14, 11:28 UTC · 35Nov 14, 13:28 UTC · 36Nov 14, 15:28 UTC · 36Nov 14, 17:28 UTC · 36Nov 14, 19:28 UTC · 36Nov 14, 21:28 UTC · 36Nov 14, 23:28 UTC · 36Nov 15, 01:28 UTC · 36Nov 15, 03:28 UTC · 37Nov 15, 05:28 UTC · 37Nov 15, 07:28 UTC · 37Nov 15, 09:28 UTC · 37Nov 15, 11:28 UTC · 37Nov 15, 13:28 UTC · 37Nov 15, 15:28 UTC · 37Nov 15, 17:28 UTC · 37Nov 15, 19:28 UTC · 37Nov 15, 21:28 UTC · 37Nov 15, 23:28 UTC · 39Nov 16, 01:28 UTC · 39Nov 16, 03:28 UTC · 39

Questions

Is SteerDistill: Quantizing LLM Agents via Closed-Loop Teacher Steering 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 SteerDistill: Quantizing LLM Agents via Closed-Loop Teacher Steering reached?

#4 on Hugging Face · Daily papers, on Nov 13, 05:28 UTC. Trendoday checks this list every 120 minutes.

How long has SteerDistill: Quantizing LLM Agents via Closed-Loop Teacher Steering been trending?

It was first seen on Nov 12, 2026 and has been on the list on 5 days (46 checks) since.