Skip to content

ci: run model-scoped pypto-serving accuracy guards #759

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
zhangqi-chen merged 1 commit into from
Jul 13, 2026
+84 −26
Merged

ci: run model-scoped pypto-serving accuracy guards #759

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
59 changes: 47 additions & 12 deletions .github/actions/pypto-serving-tests/action.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,7 +1,7 @@
name: Run pypto-serving Qwen tests
name: Run pypto-serving model tests
description: >-
Checkout pypto-serving, replace its pypto-lib submodule with the current
pypto-lib checkout under test, and run the Qwen serving tests via
pypto-lib checkout under test, and run the selected serving tests via
task-submit.

inputs:
Expand All @@ -11,12 +11,11 @@ inputs:
pypto-serving-checkout-path:
description: Workspace-relative path where pypto-serving should be checked out.
required: true
pypto-serving-ref:
description: pypto-serving ref to test against.
required: false
default: main
model-name:
description: Serving model guard to run (qwen3-14b or deepseek-v4).
required: true
model-dir:
description: Qwen3 model directory used by pypto-serving tests.
description: Model directory used by the selected pypto-serving test.
required: true
pto2-ring-dep-pool:
description: PTO2_RING_DEP_POOL value for pypto-serving tests.
Expand All @@ -38,7 +37,7 @@ runs:
uses: actions/checkout@v4
with:
repository: hw-native-sys/pypto-serving
ref: ${{ inputs.pypto-serving-ref }}
ref: main
path: ${{ inputs.pypto-serving-checkout-path }}
submodules: recursive
persist-credentials: false
Expand All @@ -61,9 +60,10 @@ runs:
working-directory: ${{ inputs.pypto-lib-checkout-path }}
run: |
source activate.sh
pip install transformers safetensors pytest
pip install transformers safetensors pytest fastapi uvicorn

- name: Run pypto-serving Qwen tests
if: ${{ inputs.model-name == 'qwen3-14b' }}
shell: bash
env:
PYPTO_QWEN3_MODEL_DIR: ${{ inputs.model-dir }}
Expand All @@ -74,14 +74,49 @@ runs:
PYPTO_SERVING_CHECKOUT: ${{ inputs.pypto-serving-checkout-path }}
run: |
source "$GITHUB_WORKSPACE/$PYPTO_LIB_CHECKOUT/activate.sh"
run_cmd="cd $GITHUB_WORKSPACE/$PYPTO_SERVING_CHECKOUT"
run_cmd="$run_cmd && source $GITHUB_WORKSPACE/$PYPTO_LIB_CHECKOUT/activate.sh"
run_cmd="source $GITHUB_WORKSPACE/$PYPTO_LIB_CHECKOUT/activate.sh"
run_cmd="$run_cmd && cd $GITHUB_WORKSPACE/$PYPTO_SERVING_CHECKOUT"
run_cmd="$run_cmd && DEVICE_ID=\$TASK_DEVICE"
run_cmd="$run_cmd PYTHONPATH=$GITHUB_WORKSPACE/$PYPTO_SERVING_CHECKOUT"
run_cmd="$run_cmd PYPTO_QWEN3_MODEL_DIR=$PYPTO_QWEN3_MODEL_DIR"
run_cmd="$run_cmd PTO2_RING_DEP_POOL=$PTO2_RING_DEP_POOL"
run_cmd="$run_cmd PTO2_RING_TASK_WINDOW=$PTO2_RING_TASK_WINDOW"
run_cmd="$run_cmd PTO2_RING_HEAP=$PTO2_RING_HEAP"
run_cmd="$run_cmd python -m pytest tests/test_qwen3_accuracy.py -q -s"
task-submit --device "$DEVICE_ID" --timeout 0 --max-time 0 \
task-submit --device "$DEVICE_ID" --timeout 1200 --max-time 1800 \
--run "$run_cmd"

- name: Run pypto-serving DeepSeek V4 accuracy test
if: ${{ inputs.model-name == 'deepseek-v4' }}
shell: bash
env:
PYPTO_DSV4_MODEL_DIR: ${{ inputs.model-dir }}
PYPTO_RUNTIME_LOG: error
PTO2_RING_DEP_POOL: ${{ inputs.pto2-ring-dep-pool }}
PTO2_RING_TASK_WINDOW: ${{ inputs.pto2-ring-task-window }}
PTO2_RING_HEAP: ${{ inputs.pto2-ring-heap }}
PTO2_OP_EXECUTE_TIMEOUT_US: 400000000
PTO2_STREAM_SYNC_TIMEOUT_MS: 440000
PTO2_SCHEDULER_TIMEOUT_MS: 320000
SERVING_WORKER_STEP_TIMEOUT: 1800
PYPTO_LIB_CHECKOUT: ${{ inputs.pypto-lib-checkout-path }}
PYPTO_SERVING_CHECKOUT: ${{ inputs.pypto-serving-checkout-path }}
run: |
source "$GITHUB_WORKSPACE/$PYPTO_LIB_CHECKOUT/activate.sh"
run_cmd="source $GITHUB_WORKSPACE/$PYPTO_LIB_CHECKOUT/activate.sh"
run_cmd="$run_cmd && cd $GITHUB_WORKSPACE/$PYPTO_SERVING_CHECKOUT"
run_cmd="$run_cmd && TASK_DEVICE=\$TASK_DEVICE"

@gemini-code-assist gemini-code-assist Bot Jul 13, 2026

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

high

There is a copy-paste error here. In the Qwen test step, DEVICE_ID is set to $TASK_DEVICE (i.e., DEVICE_ID=\$TASK_DEVICE) so that the active Ascend device ID is correctly propagated to the pytest command. Here, TASK_DEVICE=\$TASK_DEVICE is a redundant self-assignment since TASK_DEVICE is already set by task-submit. It should be changed to DEVICE_ID=\$TASK_DEVICE to ensure the correct device ID environment variable is passed to the DeepSeek test run.

        run_cmd="$run_cmd && DEVICE_ID=\\$TASK_DEVICE"

run_cmd="$run_cmd PYTHONPATH=$GITHUB_WORKSPACE/$PYPTO_SERVING_CHECKOUT"
run_cmd="$run_cmd PYPTO_DSV4_MODEL_DIR=$PYPTO_DSV4_MODEL_DIR"
run_cmd="$run_cmd PYPTO_RUNTIME_LOG=$PYPTO_RUNTIME_LOG"
run_cmd="$run_cmd PTO2_RING_DEP_POOL=$PTO2_RING_DEP_POOL"
run_cmd="$run_cmd PTO2_RING_TASK_WINDOW=$PTO2_RING_TASK_WINDOW"
run_cmd="$run_cmd PTO2_RING_HEAP=$PTO2_RING_HEAP"
run_cmd="$run_cmd PTO2_OP_EXECUTE_TIMEOUT_US=$PTO2_OP_EXECUTE_TIMEOUT_US"
run_cmd="$run_cmd PTO2_STREAM_SYNC_TIMEOUT_MS=$PTO2_STREAM_SYNC_TIMEOUT_MS"
run_cmd="$run_cmd PTO2_SCHEDULER_TIMEOUT_MS=$PTO2_SCHEDULER_TIMEOUT_MS"
run_cmd="$run_cmd SERVING_WORKER_STEP_TIMEOUT=$SERVING_WORKER_STEP_TIMEOUT"
run_cmd="$run_cmd python -m pytest tests/test_deepseek_v4_accuracy.py -q -s"
task-submit --device auto --device-num 8 --ignore-whitelist \
--timeout 1200 --max-time 1800 \
--run "$run_cmd"
51 changes: 37 additions & 14 deletions .github/workflows/ci.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -56,7 +56,8 @@ jobs:
outputs:
has_changes: ${{ steps.check.outputs.has_changes }}
files: ${{ steps.check.outputs.files }}
run_serving_tests: ${{ steps.check.outputs.run_serving_tests }}
run_qwen_serving_tests: ${{ steps.check.outputs.run_qwen_serving_tests }}
run_deepseek_serving_tests: ${{ steps.check.outputs.run_deepseek_serving_tests }}
steps:
- uses: actions/checkout@v4
with:
Expand All @@ -69,8 +70,9 @@ jobs:
if [ "${{ github.event_name }}" != "pull_request" ]; then
echo "has_changes=true" >> "$GITHUB_OUTPUT"
echo "files=" >> "$GITHUB_OUTPUT"
# serving is a PR-only job (see its `if`), so never arm it here.
echo "run_serving_tests=false" >> "$GITHUB_OUTPUT"
# Serving is PR-only (see its `if`), so never arm it here.
echo "run_qwen_serving_tests=false" >> "$GITHUB_OUTPUT"
echo "run_deepseek_serving_tests=false" >> "$GITHUB_OUTPUT"
exit 0
fi
CHANGED_FILES=$(git diff --name-only \
Expand All @@ -88,13 +90,17 @@ jobs:
echo "has_changes=false" >> "$GITHUB_OUTPUT"
fi
echo "files=$FILES" >> "$GITHUB_OUTPUT"
# pypto-serving e2e only exercises Qwen3-14B (tests/test_qwen3_accuracy.py
# against the Qwen3-14B checkpoint), so only a qwen3-14b source change can
# affect it. CI-infra edits (.github/**) must not trigger this expensive job.
# Serving e2e jobs are model-specific and expensive. CI-infra edits
# (.github/**) must not trigger them.
if printf '%s\n' "$CHANGED_FILES" | grep -qE '^models/qwen3/14b/'; then
echo "run_serving_tests=true" >> "$GITHUB_OUTPUT"
echo "run_qwen_serving_tests=true" >> "$GITHUB_OUTPUT"
else
echo "run_serving_tests=false" >> "$GITHUB_OUTPUT"
echo "run_qwen_serving_tests=false" >> "$GITHUB_OUTPUT"
fi
if printf '%s\n' "$CHANGED_FILES" | grep -qE '^models/deepseek/v4/'; then
echo "run_deepseek_serving_tests=true" >> "$GITHUB_OUTPUT"
else
echo "run_deepseek_serving_tests=false" >> "$GITHUB_OUTPUT"
fi
echo "Selected files: $FILES"

Expand Down Expand Up @@ -301,12 +307,15 @@ jobs:
working-directory: dist-checkout
run: rm -rf build_output

serving-qwen3-14b:
serving:
needs: detect-changes
# PR-only: push/dispatch never run the (expensive) qwen3-14b serving e2e.
if: github.event_name == 'pull_request' && needs.detect-changes.outputs.run_serving_tests == 'true'
# PR-only: run one shared setup, then the guards selected by model changes.
if: >-
github.event_name == 'pull_request' &&
(needs.detect-changes.outputs.run_qwen_serving_tests == 'true' ||
needs.detect-changes.outputs.run_deepseek_serving_tests == 'true')
runs-on: [self-hosted, linux, arm64, npu]
timeout-minutes: 60
timeout-minutes: 90
env:
ASCEND_HOME_PATH: /usr/local/Ascend/cann-9.0.0
PTOAS_ROOT: ${{ github.workspace }}/dist-checkout/ptoas-bin
Expand All @@ -333,13 +342,27 @@ jobs:
checkout-path: dist-checkout
build-namespace: serving

- name: Run pypto-serving e2e model tests
- name: Run pypto-serving Qwen3-14B tests
if: needs.detect-changes.outputs.run_qwen_serving_tests == 'true'
uses: ./.github/actions/pypto-serving-tests
with:
pypto-lib-checkout-path: dist-checkout
pypto-serving-checkout-path: serving-checkout
model-name: qwen3-14b
model-dir: /data/l00955553/model/Qwen3-14B

- name: Run pypto-serving DeepSeek V4 tests
if: needs.detect-changes.outputs.run_deepseek_serving_tests == 'true'
uses: ./.github/actions/pypto-serving-tests
with:
pypto-lib-checkout-path: dist-checkout
pypto-serving-checkout-path: serving-checkout
model-name: deepseek-v4
model-dir: /data/l00955553/model/dsv4-flash-w8a8
pto2-ring-dep-pool: "131072"
pto2-ring-task-window: "131072"
pto2-ring-heap: "2147483648"

- name: Kill orphaned task-submit tasks
if: always()
uses: ./.github/actions/kill-orphaned-tasks
Expand All @@ -348,4 +371,4 @@ jobs:

- name: Clean up build artifacts
if: always()
run: rm -rf dist-checkout/build_output serving-checkout/build_output serving-checkout/pypto-lib/build_output
run: rm -rf serving-checkout/build_output
Loading