CI when PR merged to main (#40451)

* up

* up

* up

* up

* up

* update

---------

Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>
This commit is contained in:
Yih-Dar
2025-08-27 10:56:18 +02:00
committed by GitHub
parent ff8b88a948
commit 80f4c0c6a0
7 changed files with 189 additions and 154 deletions

View File

@ -4,17 +4,6 @@ on:
push:
branches: [ main ]
env:
OUTPUT_SLACK_CHANNEL_ID: "C06L2SGMEEA"
HF_HUB_READ_TOKEN: ${{ secrets.HF_HUB_READ_TOKEN }}
HF_HOME: /mnt/cache
TRANSFORMERS_IS_CI: yes
OMP_NUM_THREADS: 8
MKL_NUM_THREADS: 8
RUN_SLOW: yes # For gated repositories, we still need to agree to share information on the Hub repo. page in order to get access. # This token is created under the bot `hf-transformers-bot`.
SIGOPT_API_TOKEN: ${{ secrets.SIGOPT_API_TOKEN }}
TF_FORCE_GPU_ALLOW_GROWTH: true
jobs:
get_modified_models:
name: "Get all modified files"
@ -25,111 +14,143 @@ jobs:
- name: Check out code
uses: actions/checkout@v4
- name: Get changed files
id: changed-files
uses: tj-actions/changed-files@1c8e6069583811afb28f97afeaf8e7da80c6be5c
- name: Get changed files using `actions/github-script`
id: get-changed-files
uses: actions/github-script@v7
with:
files: src/transformers/models/**
script: |
let files = [];
- name: Run step if only the files listed above change
if: steps.changed-files.outputs.any_changed == 'true'
id: set-matrix
// Only handle push events
if (context.eventName === 'push') {
const afterSha = context.payload.after;
const branchName = context.payload.ref.replace('refs/heads/', '');
let baseSha;
if (branchName === 'main') {
console.log('Push to main branch, comparing to parent commit');
// Get the parent commit of the pushed commit
const { data: commit } = await github.rest.repos.getCommit({
owner: context.repo.owner,
repo: context.repo.repo,
ref: afterSha
});
baseSha = commit.parents[0]?.sha;
if (!baseSha) {
throw new Error('No parent commit found for the pushed commit');
}
} else {
console.log(`Push to branch ${branchName}, comparing to main`);
baseSha = 'main';
}
const { data: comparison } = await github.rest.repos.compareCommits({
owner: context.repo.owner,
repo: context.repo.repo,
base: baseSha,
head: afterSha
});
// Include added, modified, and renamed files
files = comparison.files
.filter(file => file.status === 'added' || file.status === 'modified' || file.status === 'renamed')
.map(file => file.filename);
}
// Include all files under src/transformers/ (not just models subdirectory)
const filteredFiles = files.filter(file =>
file.startsWith('src/transformers/')
);
core.setOutput('changed_files', filteredFiles.join(' '));
core.setOutput('any_changed', filteredFiles.length > 0 ? 'true' : 'false');
- name: Parse changed files with Python
if: steps.get-changed-files.outputs.any_changed == 'true'
env:
ALL_CHANGED_FILES: ${{ steps.changed-files.outputs.all_changed_files }}
CHANGED_FILES: ${{ steps.get-changed-files.outputs.changed_files }}
id: set-matrix
run: |
model_arrays=()
for file in $ALL_CHANGED_FILES; do
model_path="${file#*models/}"
model_path="models/${model_path%%/*}"
if grep -qFx "$model_path" utils/important_models.txt; then
# Append the file to the matrix string
model_arrays+=("$model_path")
fi
done
matrix_string=$(printf '"%s", ' "${model_arrays[@]}" | sed 's/, $//')
echo "matrix=[$matrix_string]" >> $GITHUB_OUTPUT
test_modified_files:
python3 - << 'EOF'
import os
import sys
import json
# Add the utils directory to Python path
sys.path.insert(0, 'utils')
# Import the important models list
from important_files import IMPORTANT_MODELS
print(f"Important models: {IMPORTANT_MODELS}")
# Get the changed files from the previous step
changed_files_str = os.environ.get('CHANGED_FILES', '')
changed_files = changed_files_str.split() if changed_files_str else []
# Filter to only Python files
python_files = [f for f in changed_files if f.endswith('.py')]
print(f"Python files changed: {python_files}")
result_models = set()
# Specific files that trigger all models
transformers_utils_files = [
'modeling_utils.py',
'modeling_rope_utils.py',
'modeling_flash_attention_utils.py',
'modeling_attn_mask_utils.py',
'cache_utils.py',
'masking_utils.py',
'pytorch_utils.py'
]
# Single loop through all Python files
for file in python_files:
# Check for files under src/transformers/models/
if file.startswith('src/transformers/models/'):
remaining_path = file[len('src/transformers/models/'):]
if '/' in remaining_path:
model_dir = remaining_path.split('/')[0]
if model_dir in IMPORTANT_MODELS:
result_models.add(model_dir)
print(f"Added model directory: {model_dir}")
# Check for specific files under src/transformers/ or src/transformers/generation/ files
elif file.startswith('src/transformers/generation/') or \
(file.startswith('src/transformers/') and os.path.basename(file) in transformers_utils_files):
print(f"Found core file: {file} - including all important models")
result_models.update(IMPORTANT_MODELS)
break # No need to continue once we include all models
# Convert to sorted list and create matrix
result_list = sorted(list(result_models))
print(f"Final model list: {result_list}")
if result_list:
matrix_json = json.dumps(result_list)
print(f"matrix={matrix_json}")
# Write to GITHUB_OUTPUT
with open(os.environ['GITHUB_OUTPUT'], 'a') as f:
f.write(f"matrix={matrix_json}\n")
else:
print("matrix=[]")
with open(os.environ['GITHUB_OUTPUT'], 'a') as f:
f.write("matrix=[]\n")
EOF
model-ci:
name: Model CI
uses: ./.github/workflows/self-scheduled.yml
needs: get_modified_models
name: Slow & FA2 tests
runs-on:
group: aws-g5-4xlarge-cache
container:
image: huggingface/transformers-all-latest-gpu
options: --gpus all --privileged --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
if: ${{ needs.get_modified_models.outputs.matrix != '[]' && needs.get_modified_models.outputs.matrix != '' && fromJson(needs.get_modified_models.outputs.matrix)[0] != null }}
strategy:
fail-fast: false
matrix:
model-name: ${{ fromJson(needs.get_modified_models.outputs.matrix) }}
steps:
- name: Check out code
uses: actions/checkout@v4
- name: Install locally transformers & other libs
run: |
apt install sudo
sudo -H pip install --upgrade pip
sudo -H pip uninstall -y transformers
sudo -H pip install -U -e ".[testing]"
MAX_JOBS=4 pip install flash-attn --no-build-isolation
pip install bitsandbytes
- name: NVIDIA-SMI
run: |
nvidia-smi
- name: Show installed libraries and their versions
run: pip freeze
- name: Run FA2 tests
id: run_fa2_tests
run:
pytest -rsfE -m "flash_attn_test" --make-reports=${{ matrix.model-name }}_fa2_tests/ tests/${{ matrix.model-name }}/test_modeling_*
- name: "Test suite reports artifacts: ${{ matrix.model-name }}_fa2_tests"
if: ${{ always() }}
uses: actions/upload-artifact@v4
with:
name: ${{ matrix.model-name }}_fa2_tests
path: /transformers/reports/${{ matrix.model-name }}_fa2_tests
- name: Post to Slack
if: always()
uses: huggingface/hf-workflows/.github/actions/post-slack@main
with:
slack_channel: ${{ env.OUTPUT_SLACK_CHANNEL_ID }}
title: 🤗 Results of the FA2 tests - ${{ matrix.model-name }}
status: ${{ steps.run_fa2_tests.conclusion}}
slack_token: ${{ secrets.CI_SLACK_BOT_TOKEN }}
- name: Run integration tests
id: run_integration_tests
if: always()
run:
pytest -rsfE -k "IntegrationTest" --make-reports=tests_integration_${{ matrix.model-name }} tests/${{ matrix.model-name }}/test_modeling_*
- name: "Test suite reports artifacts: tests_integration_${{ matrix.model-name }}"
if: ${{ always() }}
uses: actions/upload-artifact@v4
with:
name: tests_integration_${{ matrix.model-name }}
path: /transformers/reports/tests_integration_${{ matrix.model-name }}
- name: Post to Slack
if: always()
uses: huggingface/hf-workflows/.github/actions/post-slack@main
with:
slack_channel: ${{ env.OUTPUT_SLACK_CHANNEL_ID }}
title: 🤗 Results of the Integration tests - ${{ matrix.model-name }}
status: ${{ steps.run_integration_tests.conclusion}}
slack_token: ${{ secrets.CI_SLACK_BOT_TOKEN }}
- name: Tailscale # In order to be able to SSH when a test fails
if: ${{ runner.debug == '1'}}
uses: huggingface/tailscale-action@v1
with:
authkey: ${{ secrets.TAILSCALE_SSH_AUTHKEY }}
slackChannel: ${{ secrets.SLACK_CIFEEDBACK_CHANNEL }}
slackToken: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}
waitForSSH: true
with:
job: run_models_gpu
slack_report_channel: "#transformers-ci-push"
docker: huggingface/transformers-all-latest-gpu
ci_event: push
report_repo_id: hf-internal-testing/transformers_ci_push
commit_sha: ${{ github.sha }}
models: ${{ needs.get_modified_models.outputs.matrix }}
secrets: inherit

View File

@ -31,7 +31,10 @@ on:
commit_sha:
required: false
type: string
models:
default: ""
required: false
type: string
env:
HF_HOME: /mnt/cache
@ -68,7 +71,7 @@ jobs:
- name: Update clone
working-directory: /transformers
run: |
git fetch && git checkout ${{ github.sha }}
git fetch && git checkout ${{ inputs.commit_sha || github.sha }}
- name: Cleanup
working-directory: /transformers
@ -87,7 +90,7 @@ jobs:
working-directory: /transformers/tests
run: |
if [ "${{ inputs.job }}" = "run_models_gpu" ]; then
echo "folder_slices=$(python3 ../utils/split_model_tests.py --num_splits ${{ env.NUM_SLICES }})" >> $GITHUB_OUTPUT
echo "folder_slices=$(python3 ../utils/split_model_tests.py --models '${{ inputs.models }}' --num_splits ${{ env.NUM_SLICES }})" >> $GITHUB_OUTPUT
echo "slice_ids=$(python3 -c 'd = list(range(${{ env.NUM_SLICES }})); print(d)')" >> $GITHUB_OUTPUT
echo "runner_map=$(python3 ../utils/get_runner_map.py)" >> $GITHUB_OUTPUT
elif [ "${{ inputs.job }}" = "run_trainer_and_fsdp_gpu" ]; then

View File

@ -75,6 +75,8 @@ jobs:
SLACK_REPORT_CHANNEL: ${{ inputs.slack_report_channel }}
ACCESS_REPO_INFO_TOKEN: ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
CI_EVENT: ${{ inputs.ci_event }}
# This `CI_TITLE` would be empty for `schedule` or `workflow_run` events.
CI_TITLE: ${{ github.event.head_commit.message }}
CI_SHA: ${{ inputs.commit_sha || github.sha }}
CI_TEST_JOB: ${{ inputs.job }}
SETUP_STATUS: ${{ inputs.setup_status }}

28
utils/important_files.py Normal file
View File

@ -0,0 +1,28 @@
# List here the models to always test.
IMPORTANT_MODELS = [
"auto",
"bert",
"gpt2",
"t5",
"modernbert",
"vit,clip",
"detr",
"table_transformer",
"got_ocr2",
"whisper",
"wav2vec2",
"qwen2_audio",
"speech_t5",
"csm",
"llama",
"gemma3",
"qwen2",
"mistral3",
"qwen2_5_vl",
"llava",
"smolvlm",
"internvl",
"gemma3n",
"gpt_oss",
"qwen2_5_omni",
]

View File

@ -1072,18 +1072,14 @@ if __name__ == "__main__":
pr_number_re = re.compile(r"\(#(\d+)\)$")
# Add Commit/PR title with a link for push CI
# (check the title in 2 env. variables - depending on the CI is triggered via `push` or `workflow_run` event)
ci_title_push = os.environ.get("CI_TITLE_PUSH")
ci_title_workflow_run = os.environ.get("CI_TITLE_WORKFLOW_RUN")
ci_title = ci_title_push if ci_title_push else ci_title_workflow_run
ci_title = os.environ.get("CI_TITLE", "")
ci_sha = os.environ.get("CI_SHA")
ci_url = None
if ci_sha:
ci_url = f"https://github.com/{repository_full_name}/commit/{ci_sha}"
if ci_title is not None:
if ci_title:
if ci_url is None:
raise ValueError(
"When a title is found (`ci_title`), it means a `push` event or a `workflow_run` even (triggered by "
@ -1112,9 +1108,9 @@ if __name__ == "__main__":
merged_by = ci_details["merged_by"]["login"]
if merged_by is None:
ci_title = f"<{ci_url}|{ci_title}>\nAuthor: {ci_author}"
ci_title = f"<{ci_url}|{ci_title}>\nAuthor: GH_{ci_author}"
else:
ci_title = f"<{ci_url}|{ci_title}>\nAuthor: {ci_author} | Merged by: {merged_by}"
ci_title = f"<{ci_url}|{ci_title}>\nAuthor: GH_{ci_author} | Merged by: GH_{merged_by}"
elif ci_sha:
ci_title = f"<{ci_url}|commit: {ci_sha}>"

View File

@ -33,11 +33,18 @@ python ../utils/split_model_tests.py --num_splits 64
"""
import argparse
import ast
import os
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument(
"--models",
type=str,
default="",
help="the list of pre-computed model names.",
)
parser.add_argument(
"--num_splits",
type=int,
@ -53,6 +60,10 @@ if __name__ == "__main__":
d1.remove("models")
d = d2 + d1
if args.models != "":
model_tests = ast.literal_eval(args.models)
d = sorted(filter(os.path.isdir, [f"models/{x}" for x in model_tests]))
num_jobs = len(d)
num_jobs_per_splits = num_jobs // args.num_splits

View File

@ -61,6 +61,9 @@ from typing import Optional, Union
from git import Repo
# List here the models not to be filtered by `filter_tests`.
from important_files import IMPORTANT_MODELS
PATH_TO_REPO = Path(__file__).parent.parent.resolve()
PATH_TO_EXAMPLES = PATH_TO_REPO / "examples"
@ -71,35 +74,6 @@ PATH_TO_TESTS = PATH_TO_REPO / "tests"
# This variable has effect only if `filter_models=False`.
NUM_MODELS_TO_TRIGGER_FULL_CI = 30
# List here the models to always test.
IMPORTANT_MODELS = [
"auto",
"bert",
"gpt2",
"t5",
"modernbert",
"vit,clip",
"detr",
"table_transformer",
"got_ocr2",
"whisper",
"wav2vec2",
"qwen2_audio",
"speech_t5",
"csm",
"llama",
"gemma3",
"qwen2",
"mistral3",
"qwen2_5_vl",
"llava",
"smolvlm",
"internvl",
"gemma3n",
"gpt_oss",
"qwen2_5_omni",
]
@contextmanager
def checkout_commit(repo: Repo, commit_id: str):