Skip to content

[Triton/Gluon] [CI] [DO NOT MERGE] ci: triton release_tmp2 + 3.8.0, with GDN block-ptr fix (#4950) #5079

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

Closed
yuyzhang512 wants to merge 5 commits into from
Closed

[Triton/Gluon] [CI] [DO NOT MERGE] ci: triton release_tmp2 + 3.8.0, with GDN block-ptr fix (#4950) #5079

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
ae8571b
[Triton] fix(gated_delta_rule): replace removed tl.make_block_ptr for…
yuyzhang512 Aug 24, 2026
3d0aa7f
ci: point Triton pypi index at release_tmp2 channel, bump triton to 3…
yuyzhang512 Aug 28, 2026
597d56f
Convert l2norm/wy_representation block ptrs; deselect known-failing t…
yuyzhang512 Aug 31, 2026
c094c6b
Merge branch 'main' into release-test/triton-tmp2-gdn-blockptr
yuyzhang512 Aug 31, 2026
5c854d6
ci(triton-test): drop test_mha_backward_with_pe deselect (passes on r…
yuyzhang512 Aug 31, 2026
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
6 changes: 3 additions & 3 deletions .github/scripts/download_triton_wheel.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9,15 +9,15 @@ python3 -m pip config set global.retries 15
python3 -m pip config set global.timeout 120

TRITON_DEFAULT_ROCM_VERSION="${TRITON_DEFAULT_ROCM_VERSION:-7.2.0}"
TRITON_INDEX_URL="https://pypi.amd.com/triton/release/rocm-${TRITON_DEFAULT_ROCM_VERSION}/simple/"
TRITON_INDEX_URL="https://pypi.amd.com/triton/release_tmp2/rocm-${TRITON_DEFAULT_ROCM_VERSION}/simple/"
ROCM_VERSION=$(dpkg -l rocm-core 2>/dev/null | awk '/^ii/{print $3}' || true)
if [[ -z "${ROCM_VERSION}" ]]; then
# RPM-based systems (e.g. rocm-core-7.2.0.70200-43.el8.x86_64 -> 7.2.0.70200)
ROCM_VERSION=$(rpm -q --queryformat '%{VERSION}' rocm-core 2>/dev/null || true)
fi
if [[ -n "${ROCM_VERSION}" ]]; then
ROCM_MAJOR_MINOR=$(echo "${ROCM_VERSION}" | cut -d. -f1,2)
TRITON_INDEX_URL="https://pypi.amd.com/triton/release/rocm-${ROCM_MAJOR_MINOR}.0/simple/"
TRITON_INDEX_URL="https://pypi.amd.com/triton/release_tmp2/rocm-${ROCM_MAJOR_MINOR}.0/simple/"
else
echo "rocm-core not found; using default ROCm version ${TRITON_DEFAULT_ROCM_VERSION}"
fi
Expand All @@ -29,7 +29,7 @@ python3 -m pip download \
--dest "${TRITON_WHEEL_DIR}" \
--index-url "${TRITON_INDEX_URL}" \
--extra-index-url https://pypi.org/simple \
"triton==3.7.0"
"triton==3.8.0"

echo "Downloading triton-kernels wheel from ${TRITON_INDEX_URL} into ${TRITON_WHEEL_DIR}"
python3 -m pip download \
Expand Down
6 changes: 3 additions & 3 deletions .github/scripts/install_triton.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -61,23 +61,23 @@ install_triton_from_wheelhouse() {
}

TRITON_DEFAULT_ROCM_VERSION="${TRITON_DEFAULT_ROCM_VERSION:-7.2.0}"
TRITON_INDEX_URL="https://pypi.amd.com/triton/release/rocm-${TRITON_DEFAULT_ROCM_VERSION}/simple/"
TRITON_INDEX_URL="https://pypi.amd.com/triton/release_tmp2/rocm-${TRITON_DEFAULT_ROCM_VERSION}/simple/"
ROCM_VERSION=$(dpkg -l rocm-core 2>/dev/null | awk '/^ii/{print $3}' || true)
if [[ -z "$ROCM_VERSION" ]]; then
# RPM-based systems (e.g. rocm-core-7.2.0.70200-43.el8.x86_64 -> 7.2.0.70200)
ROCM_VERSION=$(rpm -q --queryformat '%{VERSION}' rocm-core 2>/dev/null || true)
fi
if [[ -n "$ROCM_VERSION" ]]; then
ROCM_MAJOR_MINOR=$(echo "$ROCM_VERSION" | cut -d. -f1,2)
TRITON_INDEX_URL="https://pypi.amd.com/triton/release/rocm-${ROCM_MAJOR_MINOR}.0/simple/"
TRITON_INDEX_URL="https://pypi.amd.com/triton/release_tmp2/rocm-${ROCM_MAJOR_MINOR}.0/simple/"
else
echo "rocm-core not found; using default ROCm version ${TRITON_DEFAULT_ROCM_VERSION}"
fi

TRITON_WHEEL_DIR=${TRITON_WHEEL_DIR:-}
if ! install_triton_from_wheelhouse "${TRITON_WHEEL_DIR}"; then
echo "Installing triton from $TRITON_INDEX_URL"
python3 -m pip install --extra-index-url "$TRITON_INDEX_URL" "triton==3.7.0"
python3 -m pip install --extra-index-url "$TRITON_INDEX_URL" "triton==3.8.0"

echo "Installing triton-kernels from $TRITON_INDEX_URL"
python3 -m pip install --extra-index-url "$TRITON_INDEX_URL" "triton-kernels==1.0.0"
Expand Down
10 changes: 7 additions & 3 deletions .github/workflows/triton-test.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -260,9 +260,9 @@ jobs:
set -ex
echo "Running Triton Tests..."
docker exec -w /workspace triton_test mkdir -p test-reports
# MI35X: skip the MHA-PE backward test that fails an accuracy check on gfx950 with the coming Triton release.
# MI35X: skip the MHA-PE tests that fail an accuracy check on gfx950 with the coming Triton release.
docker exec -w /workspace triton_test pytest -v ${TRITON_TEST} \
--deselect "op_tests/triton_tests/attention/test_mha_with_pe.py::test_mha_backward_with_pe" \
--deselect "op_tests/triton_tests/attention/test_mha_with_pe.py::test_mha_varlen_with_pe[True-0.17-96-64-8-1-64-128]" \
--junitxml=test-reports/triton.xml

- name: Upload test logs
Expand Down Expand Up @@ -379,7 +379,11 @@ jobs:
set -ex
echo "Running Triton Tests..."
docker exec -w /workspace triton_test mkdir -p test-reports
docker exec -w /workspace triton_test pytest -v ${TRITON_TEST} --junitxml=test-reports/triton.xml
# MI300X: skip all shapes of the paged-prefill contexted_kv_attention tests that fail an accuracy check (and can hang the GPU) with the coming Triton release.
docker exec -w /workspace triton_test pytest -v ${TRITON_TEST} \
--deselect "op_tests/triton_tests/attention/test_pa_prefill.py::test_contexted_kv_attention" \
--deselect "op_tests/triton_tests/attention/test_chunked_pa_prefill.py::test_contexted_kv_attention" \
--junitxml=test-reports/triton.xml

- name: Upload test logs
uses: actions/upload-artifact@v4
Expand Down
Loading
Loading