Trendoday

The Surprising Effectiveness of Deleting Weights in LLM Post-Training

How little of a pretrained large language model (LLM) has to change for it to acquire new capabilities? We find that zeroing fewer than 0.05% of its weights, with no other modification, matches full f

  • Upvotes7
–Rank now
#9Peak rank
3Days on list
311Score · 7d
Nov 11, 2026First seen
Rank · 7 days

About this paper

How little of a pretrained large language model (LLM) has to change for it to acquire new capabilities? We find that zeroing fewer than 0.05% of its weights, with no other modification, matches full fine-tuning (FFT) at scale across three LLM post-training regimes: reinforcement learning with verifiable rewards, supervised fine-tuning, and on-policy distillation. Our method, Deletion Tuning, learns a binary mask over the gate projection of transformer blocks. Two practical advantages follow: the trained mask is hundreds to thousands of times smaller than the corresponding FFT checkpoint, and Deletion Tuning forgets far less, which addresses a central challenge in continual learning. Under…

Upvotes
7
Authors
6
Published
Nov 7, 2026

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

Trending history

The Surprising Effectiveness of Deleting Weights in LLM Post-Training was first seen on Hugging Face · Daily papers on Nov 11, 2026. Since then it has appeared in 22 checks on 3 different days. Its best position was #9 of 30, reached on Nov 13, 07:28 UTC. It dropped off the list after Nov 13, 07:28 UTC.

Rank history

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

#1#11#20#30 13:2823:5810:2820:5807:28 Nov 11, 13:28 UTC · #17Nov 11, 15:28 UTC · #17Nov 11, 17:28 UTC · #17Nov 11, 19:28 UTC · #15Nov 11, 21:28 UTC · #13Nov 11, 23:28 UTC · #13Nov 12, 01:28 UTC · #13Nov 12, 03:28 UTC · #12Nov 12, 05:28 UTC · #13Nov 12, 07:28 UTC · #13Nov 12, 09:28 UTC · #13Nov 12, 11:28 UTC · #12Nov 12, 13:28 UTC · #12Nov 12, 15:28 UTC · #12Nov 12, 17:28 UTC · #12Nov 12, 19:28 UTC · #12Nov 12, 21:28 UTC · #12Nov 12, 23:28 UTC · #12Nov 13, 01:28 UTC · #11Nov 13, 03:28 UTC · #11Nov 13, 05:28 UTC · #10Nov 13, 07:28 UTC · #9

Last 14 days

Best rank per day (UTC).

Nov 3today

Metric history

Platform metric as reported by Hugging Face.

071421 13:2823:5810:2820:5807:28 Nov 11, 13:28 UTC · 7Nov 11, 15:28 UTC · 9Nov 11, 17:28 UTC · 11Nov 11, 19:28 UTC · 12Nov 11, 21:28 UTC · 13Nov 11, 23:28 UTC · 13Nov 12, 01:28 UTC · 13Nov 12, 03:28 UTC · 15Nov 12, 05:28 UTC · 16Nov 12, 07:28 UTC · 17Nov 12, 09:28 UTC · 18Nov 12, 11:28 UTC · 18Nov 12, 13:28 UTC · 18Nov 12, 15:28 UTC · 18Nov 12, 17:28 UTC · 18Nov 12, 19:28 UTC · 19Nov 12, 21:28 UTC · 19Nov 12, 23:28 UTC · 19Nov 13, 01:28 UTC · 19Nov 13, 03:28 UTC · 21Nov 13, 05:28 UTC · 21Nov 13, 07:28 UTC · 21

Questions

Is The Surprising Effectiveness of Deleting Weights in LLM Post-Training trending on HF Daily Papers right now?

Not at the moment. It was last on the list on Nov 13, 07:28 UTC, after 3 days of trending.

What is the highest rank The Surprising Effectiveness of Deleting Weights in LLM Post-Training reached?

#9 on Hugging Face · Daily papers, on Nov 13, 07:28 UTC. Trendoday checks this list every 120 minutes.

How long has The Surprising Effectiveness of Deleting Weights in LLM Post-Training been trending?

It was first seen on Nov 11, 2026 and has been on the list on 3 days (22 checks) since.