DeepSpeed
DeepSpeed copied to clipboard
Fixing several issue in API and kernels to run inference with model-parallelism