Versions Compared

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.

...

Need to add amp convert_model API support for different bindings like C++, Scala etc. 

Performance

Setup

EC2 Instance: p3.8xlarge

Commit Hash: b3b952f9d5490ee2707209ab866e6c3f094e2046 (PoC changes made on top of this built from source)

Mixed Precision Models:

Resnet50_v1: JSON FileParams File

imagenet1k-resnet-152: JSON File, Params File

Results

Model (Samples/sec)Batch SizeOriginal Model (Samples/sec)Mixed Precision Model (Samples/sec)



imagenet1k-resnet-152

18572
2140

140

4240270
8320470
16405680



resnet50_v1

1215165
2370330
4560600
8760980
169351400


FAQ

Will the arg_params and aux_params be casted to fp16 ?

...