Commit History

Fix RoPE dtype cast for bfloat16 inference
9b5174f
verified

prometheus04 commited on

Add trained checkpoint: 3B tokens, loss=3.16, MFU=31.5%
5442313
verified

prometheus04 commited on

second review fixes
f4d2cf2
verified

prometheus04 commited on

Muon optimizer + README
3ac4183
verified

prometheus04 commited on

GPU-session fixes (RNG cpu, shard filter, cu124, 3090 config)
511257f
verified

prometheus04 commited on

Matilda-Mini phases 1-5 + runbook
880f286
verified

prometheus04 commited on