auto-round
feat: handle per-tensor FP8 dequantization for Devstral models
#1356
Merged
Merged