DeepSpeed
Fix a bug in the implementation of dequantization for inference
#3433
Merged
Merged