Transformer reportedly scores 76% on ARC-AGI 1 using a single H100
A user says training from scratch plus inference took a little over four hours, while a throughput-optimized recipe reached 44% in 17 minutes.
TLDR
A user reports reaching 76% on ARC-AGI 1 with a simple autoregressive transformer “with a few twists.” According to the account, training from scratch plus inference took a little over four hours on a single H100. The user also reports that a throughput-optimized recipe reached 44% in 17 minutes, describing that as TRM-level performance.
Combined views
—
1 Source, first seen ago
