Trendoday

On-Policy Morphing into Hybrid Attention Models

Hybrid attention models combine full and linear attention to achieve a better quality--efficiency trade-off. Converting pretrained Transformers into such models requires identifying which layers shoul

  • Upvotes6
#21Rank now –
#20Peak rank
3Days on list
189Score · 7d
Nov 20, 2026First seen
Rank · 7 days

About this paper

FlashMorph2 performs on-policy layer selection and weight adaptation for hybrid attention Transformers, improving long-context benchmark scores while selecting layers efficiently across model scales.

Hybrid attention models combine full and linear attention to achieve a better quality--efficiency trade-off. Converting pretrained Transformers into such models requires identifying which layers should retain full attention under a prescribed budget. We identify two off-policy mismatches in FlashMorph's layer selection: (i) continuous branch interpolation and teacher-forced hidden states differ from discrete hybrid computation; and (ii) frozen model weights fail to account for parameter adaptation during subsequent recovery. We introduce FlashMorph2, an on-policy layer-selection method that (I) evaluates sampled discrete hybrid architectures end to end using their own hidden states, and…

hybrid attention modelsfull attentionlinear attentionpretrained Transformerslayer selectionoff-policy mismatchescontinuous branch interpolationteacher-forced hidden statesdiscrete hybrid computationfrozen model weights
Upvotes
6
Authors
6
Published
Nov 18, 2026

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

Trending history

On-Policy Morphing into Hybrid Attention Models was first seen on Hugging Face · Daily papers on Nov 20, 2026. Since then it has appeared in 30 checks on 3 different days. Its best position was #20 of 30, reached on Nov 20, 17:29 UTC. It is #21 right now, holding steady at #21 over the last 6 checks.

Rank history

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

#1#11#20#30 Nov 20Nov 20Nov 21Nov 22Nov 22 Nov 20, 05:29 UTC · #24Nov 20, 07:29 UTC · #26Nov 20, 09:29 UTC · #26Nov 20, 11:29 UTC · #24Nov 20, 13:29 UTC · #25Nov 20, 15:29 UTC · #23Nov 20, 17:29 UTC · #20Nov 20, 19:29 UTC · #20Nov 20, 21:29 UTC · #21Nov 20, 23:29 UTC · #20Nov 21, 01:29 UTC · #20Nov 21, 03:29 UTC · #20Nov 21, 05:29 UTC · #20Nov 21, 07:29 UTC · #21Nov 21, 09:29 UTC · #21Nov 21, 11:29 UTC · #21Nov 21, 13:29 UTC · #21Nov 21, 15:29 UTC · #21Nov 21, 17:29 UTC · #21Nov 21, 19:29 UTC · #21Nov 21, 21:29 UTC · #21Nov 21, 23:29 UTC · #21Nov 22, 01:29 UTC · #21Nov 22, 03:29 UTC · #21Nov 22, 05:29 UTC · #21Nov 22, 07:29 UTC · #21Nov 22, 09:29 UTC · #21Nov 22, 11:29 UTC · #21Nov 22, 13:29 UTC · #21Nov 22, 15:29 UTC · #21

Last 14 days

Best rank per day (UTC).

Nov 9today

Metric history

▲ as reported by Hugging Face.

05914 Nov 20Nov 20Nov 21Nov 22Nov 22 Nov 20, 05:29 UTC · 6Nov 20, 07:29 UTC · 7Nov 20, 09:29 UTC · 7Nov 20, 11:29 UTC · 8Nov 20, 13:29 UTC · 8Nov 20, 15:29 UTC · 8Nov 20, 17:29 UTC · 10Nov 20, 19:29 UTC · 10Nov 20, 21:29 UTC · 10Nov 20, 23:29 UTC · 11Nov 21, 01:29 UTC · 12Nov 21, 03:29 UTC · 12Nov 21, 05:29 UTC · 13Nov 21, 07:29 UTC · 13Nov 21, 09:29 UTC · 13Nov 21, 11:29 UTC · 13Nov 21, 13:29 UTC · 13Nov 21, 15:29 UTC · 13Nov 21, 17:29 UTC · 13Nov 21, 19:29 UTC · 13Nov 21, 21:29 UTC · 13Nov 21, 23:29 UTC · 13Nov 22, 01:29 UTC · 13Nov 22, 03:29 UTC · 13Nov 22, 05:29 UTC · 13Nov 22, 07:29 UTC · 13Nov 22, 09:29 UTC · 13Nov 22, 11:29 UTC · 13Nov 22, 13:29 UTC · 14Nov 22, 15:29 UTC · 14

Questions

Is On-Policy Morphing into Hybrid Attention Models trending on HF Daily Papers right now?

Yes. As of Nov 22, 15:29 UTC it is #21, holding steady at #21 over the last 6 checks.

What is the highest rank On-Policy Morphing into Hybrid Attention Models reached?

#20 on Hugging Face · Daily papers, on Nov 20, 17:29 UTC. Trendoday checks this list every 120 minutes.

How long has On-Policy Morphing into Hybrid Attention Models been trending?

It was first seen on Nov 20, 2026 and has been on the list on 3 days (30 checks) since.