test(qdc): run llama_cpp LLM + VLM QDC matrix on cpu/gpu/npu - #1308
Draft
Mengsheng Wu (mengshengwu) wants to merge 2 commits into
Draft
test(qdc): run llama_cpp LLM + VLM QDC matrix on cpu/gpu/npu#1308Mengsheng Wu (mengshengwu) wants to merge 2 commits into
Mengsheng Wu (mengshengwu) wants to merge 2 commits into
Conversation
Previously the QDC llama_cpp cells only exercised cpu + npu, and test_vlm_multi_turn hard-coded cpu. Extend the backend lists to cover the OpenCL GPU path too, parametrise test_vlm_multi_turn, register the new device_gpu marker and gate it behind snapdragon so non-Snapdragon hosts still skip cleanly. Signed-off-by: Mengsheng Wu <mengshengwu@outlook.com>
Mengsheng Wu (mengshengwu)
requested a review
from Paul Zhu (vinovo)
as a code owner
August 7, 2026 14:37
The QDC harness zips the whole tests/ tree onto the device, so tests/test_*.py and tests/conftest.py changes actually affect the on-device run — but the change-gate only matched tests/qdc/**, so any pure test-code change skipped both QDC legs and shipped unverified. Widen the include to tests/. Signed-off-by: Mengsheng Wu <mengshengwu@outlook.com>
Mengsheng Wu (mengshengwu)
marked this pull request as draft
August 7, 2026 15:17
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
The QDC device matrix for the
llama_cppplugin previously exercised onlycpu+npu, andtest_vlm_multi_turnwas hard-coded tocpu. This extends both the LLM and VLM cells to also cover the OpenCLgpubackend so a regression on the llama.cpp GPU path shows up on QDC just like the NPU path does.Concretely:
tests/test_llama_cpp.py—_LLM_BACKENDS/_VLM_BACKENDSare now['cpu', 'gpu', 'npu'];test_vlm_multi_turnis parametrised ondevice_mapinstead of pinning'cpu'.tests/conftest.py—device_map='gpu'now auto-tagsdevice_gpu+snapdragon, so non-Snapdragon hosts skip the GPU cells the same way they already skip NPU.tests/pytest.ini— register the newdevice_gpumarker (--strict-markers).tests/README.md— reflect the new backend row.qairtis unchanged (stillnpu-only by design — seesdk/src/device.cpp, which coercescpu/gputo NPU with a warning).Test plan
pytest tests --collect-only -m "llama_cpp"picks up the new[gpu]cells (20 items collected on this host, 22 deselected).pytest tests --collect-only -m "api or (llama_cpp and device_cpu)"still selects only the cpu subset — the model-free + cpu shard on GitHub runners is unaffected.pytest tests --collect-only -m device_gpuyields exactly the 6 new gpu cells._qdc-test.ymlonworkflow_dispatchto confirm the new[gpu]cells run green on real QCS9075M / SC8480XP hardware.