Skip to content

Test SDPA fusion via MHA #2366

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.

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

Merged
merged 22 commits into from
Jun 5, 2025
Merged

Test SDPA fusion via MHA #2366

merged 22 commits into from
Jun 5, 2025

Conversation

gramalingam
Copy link
Collaborator

Implements SDPA (introduced by our fusions) via MHA (in a subset of cases), so that the fused model can be run and tested using ORT.

Not yet addressed: use of KV cache, 3D vs 4D Q/K/V formats. (Will address them as I cleanup the MHA fusion rules next).

Also fix some copy-paste errors in the SDPA test-cases (and make the test-case naming scheme more uniform, helps with pytest test-selection filter -k).

gramalingam and others added 22 commits May 28, 2025 12:26
Signed-off-by: Ganesan Ramalingam <grama@microsoft.com>
Signed-off-by: Ganesan Ramalingam <grama@microsoft.com>
Signed-off-by: Ganesan Ramalingam <grama@microsoft.com>
Signed-off-by: Ganesan Ramalingam <grama@microsoft.com>
Signed-off-by: Ganesan Ramalingam <grama@microsoft.com>
Signed-off-by: Ganesan Ramalingam <grama@microsoft.com>
Signed-off-by: Ganesan Ramalingam <grama@microsoft.com>
Signed-off-by: Ganesan Ramalingam <grama@microsoft.com>
Signed-off-by: Ganesan Ramalingam <grama@microsoft.com>
Signed-off-by: Ganesan Ramalingam <grama@microsoft.com>
Signed-off-by: Ganesan Ramalingam <grama@microsoft.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Copilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com>
Signed-off-by: Ganesan Ramalingam <grama@microsoft.com>
Signed-off-by: Ganesan Ramalingam <grama@microsoft.com>
Signed-off-by: Ganesan Ramalingam <grama@microsoft.com>
Signed-off-by: Ganesan Ramalingam <grama@microsoft.com>
Signed-off-by: Ganesan Ramalingam <grama@microsoft.com>
Signed-off-by: Ganesan Ramalingam <grama@microsoft.com>
Copy link

codecov bot commented Jun 4, 2025

❌ 5 Tests Failed:

Tests completed Failed Passed Skipped
16203 5 16198 2522
View the top 3 failed test(s) by shortest run time
onnxscript.backend.onnx_export_test.TestOnnxBackEnd::test_export2python_produces_correct_onnx_script_model_0577_test_if
Stack Traces | 0.004s run time
onnxscript\backend\onnx_export_test.py:137: in extract_functions
    mod = importlib.import_module(import_name)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
C:\hostedtoolcache\windows\Python\3.11.9\x64\Lib\importlib\__init__.py:126: in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
E   ModuleNotFoundError: No module named 'tests.onnx_backend_test_code.test_if'

The above exception was the direct cause of the following exception:
.nox\test_onnx_weekly\Lib\site-packages\parameterized\parameterized.py:620: in standalone_func
    return func(*(a + p.args), **p.kwargs, **kw)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
onnxscript\backend\onnx_export_test.py:271: in test_export2python_produces_correct_onnx_script_model
    functions = extract_functions(backend_test.name, code, self.test_folder)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
onnxscript\backend\onnx_export_test.py:139: in extract_functions
    raise AssertionError(
E   AssertionError: Unable to import 'tests.onnx_backend_test_code.test_if' (e=No module named 'tests.onnx_backend_test_code.test_if') (file: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_if.py', absolute path: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_if.py', current folder: D:\a\onnxscript\onnxscript
E   ---- CONTENT --
E   import numpy
E   from onnx import TensorProto
E   from onnx.helper import make_tensor
E   from onnxscript import script, external_tensor
E   from onnxscript.values import Opset
E   from onnxscript.onnx_types import BOOL, FLOAT
E   from onnxscript.onnx_opset import opset11
E   
E   @script()
E   def bck_test_if(cond: BOOL) -> (FLOAT[5]):
E       if cond:
E           then_out = opset11.Constant(value=make_tensor("value", 1, dims=[5], vals=[1.0, 2.0, 3.0, 4.0, 5.0]))
E           res = then_out
E       else:
E           else_out = opset11.Constant(value=make_tensor("value", 1, dims=[5], vals=[5.0, 4.0, 3.0, 2.0, 1.0]))
E           res = else_out
E       return res
onnxscript.backend.onnx_export_test.TestOnnxBackEnd::test_export2python_produces_correct_onnx_script_model_0866_test_pow_types_float32_int32
Stack Traces | 0.005s run time
onnxscript\backend\onnx_export_test.py:137: in extract_functions
    mod = importlib.import_module(import_name)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
C:\hostedtoolcache\windows\Python\3.11.9\x64\Lib\importlib\__init__.py:126: in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
E   ModuleNotFoundError: No module named 'tests.onnx_backend_test_code.test_pow_types_float32_int32'

The above exception was the direct cause of the following exception:
.nox\test_onnx_weekly\Lib\site-packages\parameterized\parameterized.py:620: in standalone_func
    return func(*(a + p.args), **p.kwargs, **kw)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
onnxscript\backend\onnx_export_test.py:271: in test_export2python_produces_correct_onnx_script_model
    functions = extract_functions(backend_test.name, code, self.test_folder)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
onnxscript\backend\onnx_export_test.py:139: in extract_functions
    raise AssertionError(
E   AssertionError: Unable to import 'tests.onnx_backend_test_code.test_pow_types_float32_int32' (e=No module named 'tests.onnx_backend_test_code.test_pow_types_float32_int32') (file: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_pow_types_float32_int32.py', absolute path: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_pow_types_float32_int32.py', current folder: D:\a\onnxscript\onnxscript
E   ---- CONTENT --
E   import numpy
E   from onnx import TensorProto
E   from onnx.helper import make_tensor
E   from onnxscript import script, external_tensor
E   from onnxscript.values import Opset
E   from onnxscript.onnx_types import FLOAT, INT32
E   from onnxscript.onnx_opset import opset15
E   
E   @script()
E   def bck_test_pow_types_float32_int32(x: FLOAT[3], y: INT32[3]) -> (FLOAT[3]):
E       z = opset15.Pow(x, y)
E       return z
onnxscript.backend.onnx_export_test.TestOnnxBackEnd::test_export2python_produces_correct_onnx_script_model_1269_test_unsqueeze_axis_0
Stack Traces | 0.007s run time
onnxscript\backend\onnx_export_test.py:137: in extract_functions
    mod = importlib.import_module(import_name)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
C:\hostedtoolcache\windows\Python\3.11.9\x64\Lib\importlib\__init__.py:126: in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
E   ModuleNotFoundError: No module named 'tests.onnx_backend_test_code.test_unsqueeze_axis_0'

The above exception was the direct cause of the following exception:
.nox\test_onnx_ir_git\Lib\site-packages\parameterized\parameterized.py:620: in standalone_func
    return func(*(a + p.args), **p.kwargs, **kw)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
onnxscript\backend\onnx_export_test.py:271: in test_export2python_produces_correct_onnx_script_model
    functions = extract_functions(backend_test.name, code, self.test_folder)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
onnxscript\backend\onnx_export_test.py:139: in extract_functions
    raise AssertionError(
E   AssertionError: Unable to import 'tests.onnx_backend_test_code.test_unsqueeze_axis_0' (e=No module named 'tests.onnx_backend_test_code.test_unsqueeze_axis_0') (file: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_unsqueeze_axis_0.py', absolute path: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_unsqueeze_axis_0.py', current folder: D:\a\onnxscript\onnxscript
E   ---- CONTENT --
E   import numpy
E   from onnx import TensorProto
E   from onnx.helper import make_tensor
E   from onnxscript import script, external_tensor
E   from onnxscript.values import Opset
E   from onnxscript.onnx_types import FLOAT, INT64
E   from onnxscript.onnx_opset import opset21
E   
E   @script()
E   def bck_test_unsqueeze_axis_0(x: FLOAT[3,4,5], axes: INT64[1]) -> (FLOAT[1,3,4,5]):
E       y = opset21.Unsqueeze(x, axes)
E       return y

To view more test analytics, go to the Test Analytics Dashboard
📋 Got 3 mins? Take this short survey to help us improve Test Analytics.

@github-project-automation github-project-automation bot moved this from Todo to Done in ONNX Script Review Board Jun 4, 2025
@gramalingam gramalingam merged commit c7d5786 into main Jun 5, 2025
27 of 32 checks passed
@gramalingam gramalingam deleted the rama/sdpa-via-mha branch June 5, 2025 17:01
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
Development

Successfully merging this pull request may close these issues.

2 participants