Add --durations to some pytest commands in CI workflows.

[ScottTodd]

This will help triage slow tests.

Docs:

• https://docs.pytest.org/en/stable/example/simple.html#profiling-test-duration
• https://docs.pytest.org/en/stable/how-to/usage.html#profiling-test-execution-duration

[ScottTodd]

Logs: https://github.com/nod-ai/shark-ai/actions/runs/12055697598/job/33616607357?pr=622#step:11:1630

============================= slowest 10 durations =============================
3.56s setup    tests/invocation/mobilenet_program_test.py::test_invoke_mobilenet_single_per_fiber
1.79s call     tests/invocation/mobilenet_program_test.py::test_invoke_mobilenet_parallel_per_call_explicit
1.43s setup    tests/examples/fastapi_test.py::test_error_response
1.22s call     tests/examples/async_test.py::test_async_process
0.64s setup    tests/apps/sd/components/tokenizer_test.py::test_transformers_tokenizer
0.46s call     tests/examples/async_test.py::test_async_queue
0.43s setup    tests/invocation/vmfb_buffer_access_test.py::test_kvcache_noreturn[False]
0.42s setup    tests/invocation/vmfb_buffer_access_test.py::test_kvcache_noreturn[True]
0.42s call     tests/examples/async_test.py::test_async_basic_asyncio
0.41s call     tests/invocation/vmfb_buffer_access_test.py::test_kvcache_noreturn[True]

[ScottTodd]

Logs: https://github.com/nod-ai/shark-ai/actions/runs/12055697584/job/33616607407?pr=622#step:6:201

============================= slowest 10 durations =============================
7.61s call     tests/models/llama/moe_block_test.py::MoeBlockTest::test
6.67s call     tests/examples/main_test.py::ShardingTests::testExportFfnNet
5.47s call     tests/models/punet/sharded_resnet_block_with_iree_test.py::test_sharded_resnet_block_with_iree
5.11s call     tests/models/llama/attention_test.py::AttentionBlockTest::test
3.82s call     tests/models/llama/kv_cache_test.py::KVCacheTest::testDirectAndPagedKVCachePrefill
2.48s call     tests/layers/paged_llama_attention_block_test.py::PagedLlamaAttentionBlockTest::testExportDecomposed
1.97s call     tests/kernels/mmtfp_test.py::mmtfp_test::testExportDynamicDims
1.78s call     tests/layers/paged_llama_attention_block_test.py::PagedLlamaAttentionBlockTest::testExportNondecomposed
1.55s call     tests/layers/sharded_conv2d_with_iree_test.py::test_sharded_conv2d_with_iree
1.34s call     tests/kernels/mmt_block_scaled_offset_q4_test.py::mmt_block_scaled_offset_q4_unsigned_test::testExportDynamicDims

[ScottTodd]

Logs: https://github.com/nod-ai/shark-ai/actions/runs/12055697584/job/33616608523?pr=622#step:6:547

============================== slowest durations ===============================
203.98s call     tests/models/t5/t5_test.py::T5EncoderIreeTest::testV1_1CompareXxlIreeF32AgainstTorchEagerF32
112.76s call     tests/models/t5/t5_test.py::T5EncoderIreeTest::testV1_1CompareXxlIreeBf16AgainstTorchEagerF32
26.04s call     tests/models/t5/t5_test.py::T5EncoderEagerTest::testV1_1XxlF32CompareTorchEagerAgainstHuggingFace
23.90s call     tests/models/t5/t5_test.py::T5EncoderEagerTest::testV1_1XxlBf16CompareTorchEagerAgainstHuggingFaceF32
20.66s call     tests/models/t5/t5_test.py::T5EncoderIreeTest::testV1_1CompareSmallIreeF32AgainstTorchEagerF32
19.83s call     tests/models/t5/t5_test.py::T5EncoderIreeTest::testV1_1CompareSmallIreeBf16AgainstTorchEagerF32
8.60s call     tests/models/t5/t5_test.py::T5EncoderEagerTest::testXxlBf16AgainstFluxGolden
5.76s call     tests/models/t5/t5_test.py::T5EncoderEagerTest::testV1_1SmallBf16CompareTorchEagerAgainstHuggingFace
5.21s call     tests/models/t5/t5_test.py::T5EncoderEagerTest::testV1_1SmallBf16CompareTorchEagerAgainstHuggingFaceF32
4.83s call     tests/models/t5/t5_test.py::T5EncoderEagerTest::testV1_1SmallF32CompareTorchEagerAgainstHuggingFace
2.01s call     tests/models/t5/t5_test.py::T5EncoderEagerTest::testV1_1SmallCompareTorchEagerHuggingFaceBf16AgainstF32
0.04s call     tests/models/t5/t5_test.py::T5LayerFFTest::testCompareAgainstTransformers_0
0.01s call     tests/models/t5/t5_test.py::T5LayerFFTest::testCompareAgainstTransformers_2
0.01s call     tests/models/t5/t5_test.py::T5AttentionTest::testCompareAgainstTransformers_1
0.01s call     tests/models/t5/t5_test.py::T5AttentionTest::testCompareAgainstTransformers_0

(45 durations < 0.005s hidden.  Use -vv to show these durations.)

[marbre]

Nice!

[kuhar]

@ScottTodd seems like there are unit test failures after merging this commit: https://github.com/nod-ai/shark-ai/actions/runs/12058325320/job/33624798311

[ScottTodd]

@ScottTodd seems like there are unit test failures after merging this commit: https://github.com/nod-ai/shark-ai/actions/runs/12058325320/job/33624798311

I don't see how that could be related 🤔 maybe re-run that workflow at an older commit to see if something about the deps changed?

[marbre]

I don't think it is related. It's more likely that this might be cause the latest iree nightly releases are used (?).