Skip to content

[bugfix] reset train metric when restoring a fine-tune checkpoint #2568

[bugfix] reset train metric when restoring a fine-tune checkpoint

[bugfix] reset train metric when restoring a fine-tune checkpoint #2568

Workflow file for this run

name: Unit Test CI
on:
pull_request:
types: [opened, reopened, synchronize]
workflow_dispatch:
concurrency:
group: unittest-ci-${{ github.event.pull_request.number }}
cancel-in-progress: true
jobs:
ci-test:
runs-on: tzrec-runner
container:
image: mybigpai-public-registry.cn-beijing.cr.aliyuncs.com/easyrec/tzrec-devel:1.4
options: --gpus all --ipc host --ulimit memlock=-1
steps:
- name: FetchCommit ${{ github.event.pull_request.head.sha }}
uses: actions/checkout@v4
with:
ref: ${{ github.event.pull_request.head.sha }}
path: run_${{ github.run_id }}
- name: RunUnitTestCI
id: run_unittest_ci
run: |
CACHE_ROOT="/__w/TorchEasyRec/tzrec-ci-cache"
mkdir -p "$CACHE_ROOT/triton" "$CACHE_ROOT/inductor"
cd run_${{ github.run_id }}
TRITON_CACHE_DIR="$CACHE_ROOT/triton" \
TORCHINDUCTOR_CACHE_DIR="$CACHE_ROOT/inductor" \
LD_LIBRARY_PATH=/usr/local/cuda-13/compat \
CUDA_HOME=/usr/local/cuda-13 \
bash scripts/ci/ci_test.sh --scope gpu