alphaXiv @askalphaxiv · 03 Sep 2026
“SMELT: Scaling Laws for Compute-Matched MoE Looped Transformers” Looped Transformers usually gain effective depth by spending extra compute, making it unclear whether looping itself helps. SMELT loops the middle half of an MoE Transformer twice while matching FLOPs, https://t.co/prQI7jhXHm
1 937Views
26Likes
5Reposts
1Replies
1Quotes
11Bookmarks
Is that a lot?
0.72×vs this author's median2 682 views is typical
25Percentile for this authorof 4 recent posts
0.70×vs 10K–100K median2 751 views is typical
3.59%Reachviews ÷ followers
1.70%Engagement rateof viewers reacted