THIS IS A TEST INSTANCE. ALL YOUR CHANGES WILL BE LOST!!!!
...
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 File, Params File
imagenet1k-resnet-152: JSON File, Params File
Results
Model (Samples/sec) | Batch Size | Original Model (Samples/sec) | Mixed Precision Model (Samples/sec) |
---|---|---|---|
imagenet1k-resnet-152 | 1 | 85 | 72 |
2 | 140 | 140 | |
4 | 240 | 270 | |
8 | 320 | 470 | |
16 | 405 | 680 | |
resnet50_v1 | 1 | 215 | 165 |
2 | 370 | 330 | |
4 | 560 | 600 | |
8 | 760 | 980 | |
16 | 935 | 1400 |
FAQ
Will the arg_params and aux_params be casted to fp16 ?
...