Fix failing tests with PyTorch nightly | fix(torchlib)

[justinchuby]

Follow up to fix #1061. All pipelines should now be green.

• Adjusted tolerance for float16 tests that are recently added in PyTorch
• Fix some dtype issues in function implementations

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (72b771c) 78.50% compared to head (c10e88f) 78.63%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1185      +/-   ##
==========================================
+ Coverage   78.50%   78.63%   +0.13%     
==========================================
  Files         118      118              
  Lines       15320    15323       +3     
  Branches     2388     2388              
==========================================
+ Hits        12027    12050      +23     
+ Misses       2903     2878      -25     
- Partials      390      395       +5     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[github-actions]

Test Results

         24 files  ±         0         24 suites  ±0   1h 42m 52s ⏱️ + 19m 8s
  11 212 tests +       79    8 347 ✔️ +       38      2 860 💤 +       37  5 ❌ +5 
270 080 runs  +71 411  60 699 ✔️ +11 800  209 376 💤 +59 618  5 ❌ +5 

For more details on these failures, see this check.

Results for commit c10e88f. ± Comparison against base commit 72b771c.

This pull request removes 5 and adds 84 tests. Note that renamed tests count towards both.

onnxscript.tests.function_libs.torch_lib.ops_test
onnxscript.tests.function_libs.torch_lib.ops_test.TestOutputConsistencyEagerCPU ‑ test_output_match_opinfo__nn_functional_conv3d_cpu_float32
onnxscript.tests.function_libs.torch_lib.ops_test.TestOutputConsistencyEagerCPU ‑ test_output_match_opinfo__nn_functional_conv3d_cpu_int64
onnxscript.tests.function_libs.torch_lib.ops_test.TestOutputConsistencyFullGraphCPU ‑ test_output_match_opinfo__nn_functional_conv3d_cpu_float32
onnxscript.tests.function_libs.torch_lib.ops_test.TestOutputConsistencyFullGraphCPU ‑ test_output_match_opinfo__nn_functional_conv3d_cpu_int64

onnxscript.tests.function_libs.torch_lib.ops_test.TestOutputConsistencyEagerCPU ‑ test_output_match_opinfo__addbmm_cpu_float16
onnxscript.tests.function_libs.torch_lib.ops_test.TestOutputConsistencyEagerCPU ‑ test_output_match_opinfo__addcdiv_cpu_float16
onnxscript.tests.function_libs.torch_lib.ops_test.TestOutputConsistencyEagerCPU ‑ test_output_match_opinfo__addmm_cpu_float16
onnxscript.tests.function_libs.torch_lib.ops_test.TestOutputConsistencyEagerCPU ‑ test_output_match_opinfo__addmm_decomposed_cpu_float16
onnxscript.tests.function_libs.torch_lib.ops_test.TestOutputConsistencyEagerCPU ‑ test_output_match_opinfo__addmv_cpu_float16
onnxscript.tests.function_libs.torch_lib.ops_test.TestOutputConsistencyEagerCPU ‑ test_output_match_opinfo__atan2_cpu_float16
onnxscript.tests.function_libs.torch_lib.ops_test.TestOutputConsistencyEagerCPU ‑ test_output_match_opinfo__baddbmm_cpu_float16
onnxscript.tests.function_libs.torch_lib.ops_test.TestOutputConsistencyEagerCPU ‑ test_output_match_opinfo__bernoulli_cpu_float16
onnxscript.tests.function_libs.torch_lib.ops_test.TestOutputConsistencyEagerCPU ‑ test_output_match_opinfo__bmm_cpu_float16
onnxscript.tests.function_libs.torch_lib.ops_test.TestOutputConsistencyEagerCPU ‑ test_output_match_opinfo__cross_cpu_float16
…

This pull request removes 2 skipped tests and adds 43 skipped tests. Note that renamed tests count towards both.

onnxscript.tests.function_libs.torch_lib.ops_test.TestOutputConsistencyEagerCPU ‑ test_output_match_opinfo__nn_functional_conv3d_cpu_int64
onnxscript.tests.function_libs.torch_lib.ops_test.TestOutputConsistencyFullGraphCPU ‑ test_output_match_opinfo__nn_functional_conv3d_cpu_int64

onnxscript.tests.function_libs.torch_lib.ops_test.TestOutputConsistencyEagerCPU ‑ test_output_match_opinfo__addmm_decomposed_cpu_float16
onnxscript.tests.function_libs.torch_lib.ops_test.TestOutputConsistencyEagerCPU ‑ test_output_match_opinfo__einsum_cpu_float16
onnxscript.tests.function_libs.torch_lib.ops_test.TestOutputConsistencyEagerCPU ‑ test_output_match_opinfo__linspace_tensor_overload_cpu_float16
onnxscript.tests.function_libs.torch_lib.ops_test.TestOutputConsistencyEagerCPU ‑ test_output_match_opinfo__linspace_tensor_overload_cpu_float32
onnxscript.tests.function_libs.torch_lib.ops_test.TestOutputConsistencyEagerCPU ‑ test_output_match_opinfo__linspace_tensor_overload_cpu_int32
onnxscript.tests.function_libs.torch_lib.ops_test.TestOutputConsistencyEagerCPU ‑ test_output_match_opinfo__linspace_tensor_overload_cpu_int64
onnxscript.tests.function_libs.torch_lib.ops_test.TestOutputConsistencyEagerCPU ‑ test_output_match_opinfo__log_softmax_cpu_float16
onnxscript.tests.function_libs.torch_lib.ops_test.TestOutputConsistencyEagerCPU ‑ test_output_match_opinfo__matmul_cpu_float16
onnxscript.tests.function_libs.torch_lib.ops_test.TestOutputConsistencyEagerCPU ‑ test_output_match_opinfo__nn_functional_adaptive_avg_pool1d_cpu_float16
onnxscript.tests.function_libs.torch_lib.ops_test.TestOutputConsistencyEagerCPU ‑ test_output_match_opinfo__nn_functional_adaptive_avg_pool2d_cpu_float16
…

♻️ This comment has been updated with latest results.

[BowenBao]

Changes LG, CI not green yet tho

[justinchuby]

Yes there are still a few but my vpn isn't working :(

[titaiwangms]

win pipelines are also expected to be green?

[justinchuby]

Yes. There are some other issues happening