The Surprising Effectiveness of Deleting Weights in LLM Post-Training
The paper reports that zeroing fewer than 0.05% of a pretrained LLM's weights, via a learned binary mask over the gate projection of transformer blocks, matches full fine-tuning across reinforcement learning with verifiable rewards,…
Source-based brief · Editor review not recorded · See checks and limits ↓
One primary source; other links are context, not independent corroboration.
- Primary sourcearXiv
- Discovery linkHacker News
- Linked in sourceView PDF
- Linked in sourcearXiv:2611.10866v1
- Linked in sourcecs.CL
- Linked in sourceLearn more about arXivLabs
A community post points to the original link. Discussion and upvotes do not establish accuracy.
Open full article view →- Community discussion+5
- Fetched source page+30
- No editor link review+0
- Community link+0
This measures the stored source trail and editorial steps. It does not certify accuracy, peer review, or independent corroboration. Popularity adds no points.
View full scoring →Read the scoring rubric →What the source says
The paper reports that zeroing fewer than 0.05% of a pretrained LLM's weights, via a learned binary mask over the gate projection of transformer blocks, matches full fine-tuning across reinforcement learning with verifiable rewards, supervised fine-tuning, and on-policy distillation. The method, Deletion Tuning, produces masks hundreds to thousands of times smaller than full fine-tuning checkpoints. Under reinforcement learning, it forgets less prior-task accuracy than full fine-tuning, LoRA, or random sparse fine-tuning. The authors analyze weight-change geometry using stable rank, spectrum drift, and principal-weight overlap.
Claim explorer
Attribution, source and conditionsSource-stated points from the stored brief. Claim-level assessments have not been recorded.
Source claims not independently checked; this is a preprint abstract and the full paper was not reviewed.
This connects a source, an attributed statement and its qualifications. It does not establish that the claim is correct.
Questions worth asking
Open questions for further reportingWhat remains unconfirmed?
Source claims not independently checked; this is a preprint abstract and the full paper was not reviewed.
What material was available for this brief?
The recorded brief basis is fetched source page. These points are AINewsIQ paraphrases, not quotations or independent confirmations.
Zeroing fewer than 0.05% of weights, with no other modification, reportedly matches full fine-tuning at scale.
Source: arXiv ↗Stored source-based brief · key point 1 · AINewsIQ paraphraseThe learned binary mask is hundreds to thousands of times smaller than the corresponding full fine-tuning checkpoint.
Source: arXiv ↗Stored source-based brief · key point 2 · AINewsIQ paraphraseUnder reinforcement learning, prior-task accuracy stays near base level while full fine-tuning, LoRA, and random sparse tuning lose up to 5.4 points.
Source: arXiv ↗Stored source-based brief · key point 3 · AINewsIQ paraphraseLimits from the stored brief
Source claims not independently checked; this is a preprint abstract and the full paper was not reviewed.
This is a source-based abstract, not independent validation of the source's claims.
Evidence & sources
These links show where the signal came from and what the source page linked to. A linked page is context, not independent corroboration.
A community post points to the original link. Discussion and upvotes do not establish accuracy.
Timeline & updates
- Signal dated by source or discovery feed
This timestamp places the signal in the radar. It may be the discovery post date rather than the primary page's first publication date.
- Shared brief stored
AINewsIQ prepared one abstract from fetched source page.
Brief history · 1 retained version
These are changes to AINewsIQ’s own publication. Earlier history is unavailable unless recorded here. Complete retained versions are available to the editor.
- Version 1 · published
Source-based brief stored or updated. This does not establish that the publisher changed its page.
Automated record
Source observations · 1 recent captures
A changed capture means the extracted text differed. It can reflect page furniture or extraction changes; it does not establish a substantive correction. Source captures remain private.
Retained capture; no earlier comparison available.
New engagement snapshots do not mean the source itself changed. Source observations and AINewsIQ brief revisions are separate records.
Why this may matter
If the results hold, weight masking could offer a compact alternative to full fine-tuning with less forgetting, a central problem in continual learning. Results are limited to the reported regimes and models.
Limits and open questions
Source claims not independently checked; this is a preprint abstract and the full paper was not reviewed.
Organization-specific financial, security, and operational effects require the organization's own context and review.

Join the discussion
Have a different perspective or a question about the evidence? Discussion is free with a verified email account. Every comment is reviewed before publication.
No published comments yet.
New commenters: 2 submissions per 24 hours, at least 5 minutes apart, until 3 are approved. Discussion guidelines and moderation →