Currently, we skip benchmarks for dependency updates, but since different jax versions can have significant performance differences, I believe we shouldn't. This is probably as easy as copying the lines below after Checkout benchmarks from PR head step.
- name: Restore Python environment cache
if: env.has_changes == 'true'
id: restore-env-2
uses: actions/cache/restore@v5
with:
path: .venv-${{ env.version }}
key: ${{ runner.os }}-venv-${{ env.version }}-${{ hashFiles('devtools/dev-requirements.txt', 'requirements.txt') }}
- name: Set up virtual environment if not restored from cache
if: steps.restore-env-2.outputs.cache-hit != 'true' && env.has_changes == 'true'
run: |
gh cache list
# remove previous venv folder and deactivate venv before
python -m venv .venv-${{ env.version }}
source .venv-${{ env.version }}/bin/activate
python -m pip install --upgrade pip
pip install -r devtools/dev-requirements.txt
pip install matplotlib==3.10.8
- name: Action Details
if: env.has_changes == 'true'
run: |
source .venv-${{ env.version }}/bin/activate
which python
python --version
pwd
lscpu
pip list
We also need to add requirements.txt to the Filter Changes step. Should be easy to add.
Currently, we skip benchmarks for dependency updates, but since different
jaxversions can have significant performance differences, I believe we shouldn't. This is probably as easy as copying the lines below afterCheckout benchmarks from PR headstep.We also need to add
requirements.txtto theFilter Changesstep. Should be easy to add.