| 文件 | 最后提交记录 | 最后更新时间 |
|---|---|---|
ci: refactor v2 test pipeline to use upstream run_test.py - Replace self-built run_npu_test_file.py (857 lines) with upstream pytorch/test/run_test.py + test-npu.sh entry point (187 lines) - Remove _test-collect-new.yml: config job now runs on ubuntu-latest (no NPU needed), only reads shard counts from YAML - Add fcntl-based NPU device binding in npu_poisoning_plugin.py - Add generate_shard_jsonl.py: parse JUnit XML + run_test.py stderr into JSONL format (replaces parse_test_results.py) - Switch report format from JSON to JSONL throughout the pipeline - Fix distributed.devices_per_proc: 1 -> 8 (was breaking distributed tests) - Fix NUM_PROCS: now calculated as npu_count / devices_per_proc - Set cancel-in-progress: true to avoid queued run pileup - Restore upstream_ref to main (was pinned to a fixed SHA) - Remove push trigger from nightly.yml (avoid conflict with v2) - Add CANN log collection and failure diagnostics (npu-smi, dmesg) - Add generate_test_times.py for future shard optimization - Simplify report workflow: single artifact download + auto-discover - Net: -2488 lines, +428 lines (~80% reduction) | 20 天前 | |
fix: scan run_test.py python-pytest XML output for case results run_test.py writes JUnit XMLs to {TEST_DIR}/test-reports/python-pytest/ which was not being scanned by find_xml_files, resulting in 0 XML files found and missing case-level results. Add --pytest-xml-dir argument to generate_shard_jsonl.py and pass the path from test-npu.sh. | 19 天前 | |
ci: refactor v2 test pipeline to use upstream run_test.py - Replace self-built run_npu_test_file.py (857 lines) with upstream pytorch/test/run_test.py + test-npu.sh entry point (187 lines) - Remove _test-collect-new.yml: config job now runs on ubuntu-latest (no NPU needed), only reads shard counts from YAML - Add fcntl-based NPU device binding in npu_poisoning_plugin.py - Add generate_shard_jsonl.py: parse JUnit XML + run_test.py stderr into JSONL format (replaces parse_test_results.py) - Switch report format from JSON to JSONL throughout the pipeline - Fix distributed.devices_per_proc: 1 -> 8 (was breaking distributed tests) - Fix NUM_PROCS: now calculated as npu_count / devices_per_proc - Set cancel-in-progress: true to avoid queued run pileup - Restore upstream_ref to main (was pinned to a fixed SHA) - Remove push trigger from nightly.yml (avoid conflict with v2) - Add CANN log collection and failure diagnostics (npu-smi, dmesg) - Add generate_test_times.py for future shard optimization - Simplify report workflow: single artifact download + auto-discover - Net: -2488 lines, +428 lines (~80% reduction) | 20 天前 | |
ci: replace upstream Pool with self-managed per-file scheduler - Add npu_scheduler.py: ThreadPoolExecutor + DevicePool (acquire/release model, no wrap-around conflict), per-file run_test.py -i invocation, subprocess timeout (3600s), real-time progress output, sys.exit(1) on failure so CI step turns red - test-npu.sh: call npu_scheduler.py instead of single run_test.py; capture exit code via $?; add pytest-timeout plugin (600s/case) - npu_poisoning_plugin.py: remove fcntl device binding (moved to scheduler), pytest_configure now only logs visible devices, pytest_unconfigure no longer forces os._exit(70) - _test-exec-new.yml: switch to heredoc << 'PYEOF' to avoid bash quote-escaping issues with ${{ }} expressions | 19 天前 | |
ci: replace upstream Pool with self-managed per-file scheduler - Add npu_scheduler.py: ThreadPoolExecutor + DevicePool (acquire/release model, no wrap-around conflict), per-file run_test.py -i invocation, subprocess timeout (3600s), real-time progress output, sys.exit(1) on failure so CI step turns red - test-npu.sh: call npu_scheduler.py instead of single run_test.py; capture exit code via $?; add pytest-timeout plugin (600s/case) - npu_poisoning_plugin.py: remove fcntl device binding (moved to scheduler), pytest_configure now only logs visible devices, pytest_unconfigure no longer forces os._exit(70) - _test-exec-new.yml: switch to heredoc << 'PYEOF' to avoid bash quote-escaping issues with ${{ }} expressions | 19 天前 | |
ci: exclude test/cpython from scanning to avoid import hang cpython tests (test_ordered_dict.py, test_userdict.py, etc.) have a RedirectImportFinder that causes infinite recursion during import in the NPU test environment (Python 3.10). This hangs pytest collection for 1800s per attempt, and --scs recovery cannot help because lastrun is never set (crash happens before any test runs). Add top-level 'exclude' config field to skip directories entirely during scanning — excluded files never enter any category, not even others. Remove test/cpython from graph.paths and add it to exclude. | 22 天前 | |
ci: replace upstream Pool with self-managed per-file scheduler - Add npu_scheduler.py: ThreadPoolExecutor + DevicePool (acquire/release model, no wrap-around conflict), per-file run_test.py -i invocation, subprocess timeout (3600s), real-time progress output, sys.exit(1) on failure so CI step turns red - test-npu.sh: call npu_scheduler.py instead of single run_test.py; capture exit code via $?; add pytest-timeout plugin (600s/case) - npu_poisoning_plugin.py: remove fcntl device binding (moved to scheduler), pytest_configure now only logs visible devices, pytest_unconfigure no longer forces os._exit(70) - _test-exec-new.yml: switch to heredoc << 'PYEOF' to avoid bash quote-escaping issues with ${{ }} expressions | 19 天前 |
| 文件 | 最后提交记录 | 最后更新时间 |
|---|---|---|
| 20 天前 | ||
| 19 天前 | ||
| 20 天前 | ||
| 19 天前 | ||
| 19 天前 | ||
| 22 天前 | ||
| 19 天前 |