Commit Graph

6 Commits

Author SHA1 Message Date
araleza
cd239f0fa9 Moved kahan state from file globals to optimizer state variables 2025-08-20 16:42:15 +01:00
araleza
3f0230a286 Now sending int16s instead of f32s to cpu device; faster and maybe more accurate 2025-07-29 10:05:06 +01:00
araleza
bb7750fbca Fixed typo in comment 2025-07-23 15:10:57 +01:00
araleza
6517b2b838 Added support for Kahan summation for Adafactor-optimized Flux FFT 2025-07-23 14:34:32 +01:00
Kohya S
e1cd19c0c0 add stochastic rounding, fix single block 2024-08-21 21:04:10 +09:00
2kpr
4f203ce40d Fused backward pass 2024-04-14 09:56:58 -05:00