Users praise the MuP-scaled Short Conv Networks as a beautiful architecture with competitive small-model performance and a safe base for tinkering.
Based on 4 visible X reactions from 10 accounts; directional sample.
Ask a question below.
Published answers will appear here.
it's really cool how the small one trades blows so evenly with the big one
11:49 AM · Jul 15, 2026overall pretty safe release, great base for people to use tinker on, congrats on the release!
11:51 AM · Jul 15, 2026Aggressive kernel fusion and Triton MXFP8 quantization boosted training performance.
Users praise the MuP-scaled Short Conv Networks as a beautiful architecture with competitive small-model performance and a safe base for tinkering.
Based on 4 visible X reactions from 10 accounts; directional sample.
Ask a question below.
Published answers will appear here.