fix

Author: xadupre

_scripts/test_backend_onnxruntime.py

@@ -141,7 +141,7 @@ def run_node(cls, node, inputs, device=None, outputs_info=None, **kwargs):
 
 backend_test.exclude("(test_adagrad|test_adam|test_add_uint8)")
 
-if pv.Version(onnxruntime.__version__) <= pv.Version("1.25"):
+if pv.Version(onnxruntime.__version__) <= pv.Version("1.30"):
     backend_test.exclude("(test_attention_4d_with|test_attention_4d_gqa)")
 
 

_unittests/ut_reference/test_backend_onnxruntime_evaluator.py

@@ -299,7 +299,7 @@ def run_node(cls, node, inputs, device=None, outputs_info=None, **kwargs):
     )
 
 
-if pv.Version(onnxruntime.__version__) <= pv.Version("1.25"):
+if pv.Version(onnxruntime.__version__) <= pv.Version("1.30"):
     backend_test.exclude("(test_attention_4d_with|test_attention_4d_gqa)")
 
 # import all test cases at global scope to make them visible to python.unittest

_unittests/ut_torch_export_patches/test_patch_transformers.py

@@ -645,7 +645,7 @@ def test_plug_multi_head_attention_qwen25_packed_float16(self):
             self.assertEqualArray(results.eager_outputs[0], results.onnx_outputs[0], atol=0.01)
             self.assertLess(results.diffs[0]["abs"], 0.01)
 
-    @requires_onnxruntime("1.25")
+    @requires_onnxruntime("1.30")
     @unittest.skipIf(not patch_qwen2_5, "Qwen25 not part of this transformers")
     def test_plug_multi_head_attention_qwen25_loopmha_float16(self):
         from onnx_diagnostic.torch_export_patches.patches._patch_transformers_qwen2_5 import (
@@ -681,7 +681,7 @@ def test_plug_multi_head_attention_qwen25_loopmha_float16(self):
             self.assertEqualArray(results.eager_outputs[0], results.onnx_outputs[0], atol=0.01)
             self.assertLess(results.diffs[0]["abs"], 0.01)
 
-    @requires_onnxruntime("1.25")
+    @requires_onnxruntime("1.30")
     @unittest.skipIf(not patch_qwen2_5, "Qwen25 not part of this transformers")
     def test_plug_multi_head_attention_qwen25_loopmha_float32(self):
         from onnx_diagnostic.torch_export_patches.patches._patch_transformers_qwen2_5 import (
@@ -717,7 +717,7 @@ def test_plug_multi_head_attention_qwen25_loopmha_float32(self):
             self.assertEqualArray(results.eager_outputs[0], results.onnx_outputs[0], atol=1e-5)
             self.assertLess(results.diffs[0]["abs"], 1e-5)
 
-    @requires_onnxruntime("1.25")
+    @requires_onnxruntime("1.30")
     @unittest.skipIf(not patch_qwen2_5, "Qwen25 not part of this transformers")
     def test_plug_multi_head_attention_qwen25_loopa24_float16(self):
         from onnx_diagnostic.torch_export_patches.patches._patch_transformers_qwen2_5 import (
@@ -748,7 +748,7 @@ def test_plug_multi_head_attention_qwen25_loopa24_float16(self):
             )
             self.assertLess(results.diffs[0]["abs"], 0.005)
 
-    @requires_onnxruntime("1.25")
+    @requires_onnxruntime("1.30")
     @unittest.skipIf(not patch_qwen2_5, "Qwen25 not part of this transformers")
     def test_plug_multi_head_attention_qwen25_loopa24_float32(self):
         from onnx_diagnostic.torch_export_patches.patches._patch_transformers_qwen2_5 import (

_unittests/ut_torch_onnx/test_discrepancies.py

@@ -46,7 +46,7 @@ def qwen_sdpa_attention(
             return attn_output
 
         for model_name in ["attention_loopa24.onnx", "attention_loopmha.onnx"]:
-            if model_name == "attention_loopa24.onnx" and not has_onnxruntime("1.25"):
+            if model_name == "attention_loopa24.onnx" and not has_onnxruntime("1.30"):
                 # not available
                 continue
             with self.subTest(model=model_name):