dlperf

DLPerf

Download Models

Download the models listed models in model_urls to model_dir using

go run main.go downloadmodels --model_dir ~/onnx_models

Layer Stats

model_path and output_path can be a folder or a file.

Infer and view the shape information using

go run main.go layerstats --model_path ~/onnx_models/emotion_ferplus/model.onnx --format dot

Get layer statistics in json format using

go run main.go layerstats --model_path ~/onnx_models --output_path assets/layer_stats --format json

dot from graphviz is needed. On macos, install it using brew install graphviz.

Get Flops

Get the flops information for alexnet using

go run main.go flopsinfo --model_path ~/onnx_models/bvlc_alexnet/model.onnx

Get information per layer using

go run main.go flopsinfo --model_path ~/onnx_models/bvlc_alexnet/model.onnx --full

Get Weights Histogram

Get information per layer using

go run main.go weightsinfo --model_path ~/onnx_models/bvlc_alexnet/model.onnx --output_file=out

Store the recalled benchmark in json

Store the recalled benchmarks in json using

go run main.go benchinfo --model_path ~/data/carml/ --benchmark_database results/v100/8.json --short=false --batch_size=8 --human=true -o assets/benchinfo/v100 -f json

Find Patterns

model_path and output_path can be a folder or a file.

Find the patterns of length 4

go run main.go patterns --model_path ~/onnx_models/ --length 4

Generate Benchmarks

Generate the benchmark files of a model or across models at model_path. Use --forward and --backward to control whether to generate benchmarks for forward and backward pass.

go run main.go benchgen --model_path ~/onnx_models/bvlc_alexnet/model.onnx --forward=true --backward=false -o generated_benchmarks.hpp

Query benchmark database

Query benchmark database at benchmark_database to to get information on the model at model_path

go run main.go benchinfo --model_path ~/onnx_models/vgg19/vgg19.onnx

or

go run main.go benchinfo --model_path ~/onnx_models/resnet100/resnet100.onnx --benchmark_database ../microbench/results/cudnn/ip-172-31-26-89.json

Draw a graph of the layers using benchmark data

You can draw a graph with the runtime data using the following command

go run main.go benchinfo --model_path ~/data/carml/dlperf/ArcFace/resnet100/resnet100.onnx --benchmark_database results/v100/8.json --short=false --batch_size=8 --human=true --strategy=parallel --show