CI workflow and output fixes (#189)

Author: jiannanWang

.github/workflows/failure-test.yml

@@ -14,7 +14,7 @@ jobs:
       - name: Install uv
         uses: astral-sh/setup-uv@v3
       - name: Set up Python
-        run: uv python install
+        run: uv python install 3.13
       - name: Install package and dependencies
         run: uv sync --dev
       - name: Run smoke test
@@ -37,7 +37,7 @@ jobs:
       - name: Install uv
         uses: astral-sh/setup-uv@v3
       - name: Set up Python
-        run: uv python install
+        run: uv python install 3.13
       - name: Install package and dependencies
         run: uv sync --dev
       - name: Run opinfo test
@@ -60,7 +60,7 @@ jobs:
       - name: Install uv
         uses: astral-sh/setup-uv@v3
       - name: Set up Python
-        run: uv python install
+        run: uv python install 3.13
       - name: Install package and dependencies
         run: uv sync --dev
       - name: Clone FACTO source

.github/workflows/ruff.yml

@@ -17,7 +17,7 @@ jobs:
       uses: astral-sh/setup-uv@v3
 
     - name: Set up Python
-      run: uv python install
+      run: uv python install 3.13
 
     - name: Install package with dev dependencies
       run: uv sync --dev

.github/workflows/smoke-test.yml

@@ -19,7 +19,7 @@ jobs:
       uses: astral-sh/setup-uv@v3
 
     - name: Set up Python
-      run: uv python install
+      run: uv python install 3.13
 
     - name: Install package and dependencies
       run: uv sync --dev

BackendBench/output.py

@@ -219,11 +219,11 @@ def _get_summary_op_results(
     for op in op_names:
         if len(correctness_results_dict[op]) > 0:
             correctness = sum(correctness_results_dict[op]) / len(correctness_results_dict[op])
-            correctness = f"{correctness:.4f}%"
+            correctness = f"{correctness * 100:.4f}%"
         else:
             correctness = "N/A"
         if len(speedups_dict[op]) > 0:
-            speedup = sum(speedups_dict[op]) / len(speedups_dict[op])
+            speedup = torch.tensor(speedups_dict[op]).log().mean().exp()
             speedup = f"{speedup:.4f}x"
         else:
             speedup = "N/A"

test/test_output.py

@@ -144,7 +144,7 @@ def test_get_summary_op_results(self):
         # Check that results are sorted properly (by speedup descending, then correctness)
         assert_expected_inline(
             str(op_results),
-            """[('torch.ops.aten.add.Tensor', '1.0000%', '1.7500x'), ('torch.ops.aten.sin.default', '1.0000%', '1.0000x'), ('torch.ops.aten.mul.Tensor', '0.0000%', '1.0000x')]""",
+            """[('torch.ops.aten.add.Tensor', '100.0000%', '1.7321x'), ('torch.ops.aten.sin.default', '100.0000%', '1.0000x'), ('torch.ops.aten.mul.Tensor', '0.0000%', '1.0000x')]""",
         )
 
     def test_save_results_integration(self):