/AI2h ago

Fable Shows 15% Compute Wins on NanoGPT Speedrun via PyTorch Pipeline

115011.1K
bilal#1366
Original post
bilal@bilaltwovec#1366inAI

fable not sandbagging on showing 15% compute multiplier wins on the nanogpt speedrun is the new you know its over when a gdm paper makes it past internal review to hit arxiv

12:08 PM · Jun 9, 2026 · 15 Views
Sentiment

Users accused the new Muon variant of sandbagging its claimed 15% compute gains on NanoGPT pretraining while calling the model's responses lazier and worse than prior versions.

Pos
0.0%
Neg
100.0%
2 comments with sentiment.
Cluster Engagement
Posts from X
Most Activity
Most Activity
VIEWS1.1KBOOKMARKS1LIKES15REPLIES1
bilal@bilaltwovec

fable not sandbagging on showing 15% compute multiplier wins on the nanogpt speedrun for a new muon variant is the new you know its over when a gdm paper makes it past internal review to hit arxiv

2hViews 1.1KLikes 15Bookmarks 1
bilal@bilaltwovec

omg the sandbagging is real

2hViews 72Likes 1
bilal@bilaltwovec

turned off history and tried again real quick its so much lazier and worse of a response than even 4.8 max from last night

2hViews 27