forked from pytorch/benchmark
-
Notifications
You must be signed in to change notification settings - Fork 0
66 lines (65 loc) · 2.6 KB
/
tritonbench-nightly.yml
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
name: TritonBench nightly (A100)
on:
workflow_dispatch:
schedule:
- cron: '00 18 * * *' # run at 6:00 PM UTC, K8s containers will roll out at 12PM EST
jobs:
run-benchmark:
environment: docker-s3-upload
env:
BASE_CONDA_ENV: "torchbench"
CONDA_ENV: "tritonbench-nightly"
PLATFORM_NAME: "gcp_a100"
SETUP_SCRIPT: "/workspace/setup_instance.sh"
TORCHBENCH_USERBENCHMARK_SCRIBE_GRAPHQL_ACCESS_TOKEN: ${{ secrets.TORCHBENCH_USERBENCHMARK_SCRIBE_GRAPHQL_ACCESS_TOKEN }}
AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_ID }}
AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
IS_GHA: 1
BUILD_ENVIRONMENT: benchmark-nightly
if: ${{ github.repository_owner == 'pytorch' }}
runs-on: [a100-runner]
steps:
- name: Checkout TorchBench
uses: actions/checkout@v3
with:
path: benchmark
- name: Tune Nvidia GPU
run: |
sudo nvidia-smi -pm 1
sudo nvidia-smi -ac 1215,1410
nvidia-smi
sudo ldconfig
- name: Clone and setup conda env
run: |
CONDA_ENV=${BASE_CONDA_ENV} . "${SETUP_SCRIPT}"
conda create --name "${CONDA_ENV}" --clone "${BASE_CONDA_ENV}"
- name: Run the triton userbenchmark ci
run: |
. "${SETUP_SCRIPT}"
# remove old results if any
if [ -d benchmark-output ]; then rm -Rf benchmark-output; fi
pushd benchmark
if [ -d .userbenchmark ]; then rm -Rf .userbenchmark; fi
python run_benchmark.py triton --ci
cp -r ./.userbenchmark/triton ../benchmark-output
- name: Copy artifact and upload to scribe and Amazon S3
run: |
. "${SETUP_SCRIPT}"
pushd benchmark
LATEST_RESULT=$(find ../benchmark-output/ -name "metrics-*.json" | sort -r | head -1)
echo "Benchmark result file: ${LATEST_RESULT}"
# Upload the result json to Scribe
python ./scripts/userbenchmark/upload_scribe.py --userbenchmark_json "${LATEST_RESULT}" --userbenchmark_platform "${PLATFORM_NAME}"
# Upload the result json to Amazon S3
python ./scripts/userbenchmark/upload_s3.py --upload-file "${LATEST_RESULT}" --userbenchmark_platform "${PLATFORM_NAME}"
- name: Upload result to GH Actions Artifact
uses: actions/upload-artifact@v4
with:
name: TritonBench result
path: benchmark-output/
- name: Clean up Conda env
if: always()
run: |
. "${SETUP_SCRIPT}"
conda deactivate && conda deactivate
conda remove -n "${CONDA_ENV}" --all