Skip to content

integration_tests

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History
 
 

integration_tests

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
models
models
 
 
python
python
 
 
.gitignore
.gitignore
 
 
README.md
README.md
 
 
batchnormalization3d_benchmark.ts
batchnormalization3d_benchmark.ts
 
 
benchmark_cloud.sh
benchmark_cloud.sh
 
 
benchmark_layers.sh
benchmark_layers.sh
 
 
benchmark_models_test.ts
benchmark_models_test.ts
 
 
benchmark_ops_test.ts
benchmark_ops_test.ts
 
 
benchmarks.sh
benchmarks.sh
 
 
cocossd_benchmarks.ts
cocossd_benchmarks.ts
 
 
conv_benchmarks.ts
conv_benchmarks.ts
 
 
firebase.ts
firebase.ts
 
 
firebase_types.ts
firebase_types.ts
 
 
firestore.ts
firestore.ts
 
 
karma.conf.js
karma.conf.js
 
 
karma.conf.layers.js
karma.conf.layers.js
 
 
karma.conf.validations.js
karma.conf.validations.js
 
 
matmul_benchmarks.ts
matmul_benchmarks.ts
 
 
mobilenet_benchmarks.ts
mobilenet_benchmarks.ts
 
 
package.json
package.json
 
 
run_node_tests.ts
run_node_tests.ts
 
 
test_util.ts
test_util.ts
 
 
tsconfig.json
tsconfig.json
 
 
types.ts
types.ts
 
 
use_benchmarks.ts
use_benchmarks.ts
 
 
util.ts
util.ts
 
 
validate_converter.sh
validate_converter.sh
 
 
yarn.lock
yarn.lock
 
 

README.md

TensorFlow.js Integration tests

This directory includes performance benchmarks and validation tests of TensorFlow.js. This is work in progress and the command lines and APIs are subject to frequent changes.

Running Core Benchmarks

TODO(annyuan): Add doc.

Running Layers Benchmarks in the Browser

To run the browser-based layers benchmarks, do:

yarn benchmark --layers

Use the --log flag to cause the benchmark data and related metadata to be logged to Cloud Firestore.

yarn benchmark --layers --log

Running tfjs-node Benchmarks

To run the tfjs-node (CPU) benchmarks, do:

yarn benchmark --layers --tfjs-node

To run the tfjs-node-gpu (CUDA GPU) benchmarks, do:

yarn benchmark --layers --tfjs-node-gpu

Obviously, this requires a CUDA-enabled GPU and all required drivers and libraries to be set up properly on the system.

Also, note that using the --tfjs-node-gpu will cause the the GPU (CUDA) version of TensorFlow (Python) to be installed and used for comparison.

Add the --log flag to cause the benchmark data and related metadata to be logged to Cloud Firestore. This requires you set the GOOGLE_APPLICATION_CREDENTIALS environment variable and point it to the service-account JSON file. See this Google Cloud documentation page for more details: https://cloud.google.com/docs/authentication/getting-started

Running GraphModel converter validation tests

To run the browser-based GraphModel converter validation tests, do:

yarn validate_converter