Maxim Krivchikov
Results
1
comments of
Maxim Krivchikov
Is it possible to implement saving in F16 and F32 too in this PR (dequantization)? This feature will be useful to train LoRA for quantized models.