Execution Trace Replay (et_replay)

et_replay is a tool designed for replaying Chakra Execution Traces (ET) from machine learning models.

Installation

To install et_replay, use the following commands:

$ git clone https://github.com/meta-pytorch/chakra_replay/
$ conda create -n et_replay python=3.10
$ conda activate et_replay
$ cd chakra_replay
$ pip3 install -r requirements.txt
$ pip3 install .

Running et_replay

Unzip tests/inputs/resnet_et.json.gz

gzip -d tests/inputs/resnet_et.json.gz

Replay it with the following command.

$ python3 -m et_replay.tools.et_replay --input tests/inputs/resnet_et.json -c --profile-replay

Note: When analyzing performance values from et_replay, refer to the collected Kineto traces rather than the execution time reported by et_replay. Kineto traces are only collected when --profile-replay is provided.

License

Chakra replay is released under Apache-2.0 license. Please see the LICENSE file for more information.