ci: remove GitHub Actions unit-test workflow

Author: mark14wu

.github/workflows/python-app.yml

@@ -5,9 +5,9 @@ name: Python application
 
 on:
   push:
-    branches: [ "main", "keren/v2.0" ]
+    branches: []
   pull_request:
-     branches: [ "main", "keren/v2.0" ]
+    branches: []
 
 permissions:
   contents: read

tests/test_autotune_add.py

@@ -37,8 +37,8 @@ def test_autotune_add_inrange():
     This test uses n_elements = 128, matching the size of the input tensors.
     It should NOT cause any out-of-bound access.
     """
-    x = torch.randn(128)
-    y = torch.randn(128)
+    x = torch.randn(128, device='cuda')
+    y = torch.randn(128, device='cuda')
     out = torch.empty_like(x)
 
     # The kernel launch uses n_elements=128, aligned with the tensor size.
@@ -57,8 +57,8 @@ def test_autotune_add_out_of_bound():
     This test deliberately sets n_elements = 256, exceeding the actual buffer size (128).
     It will likely cause out-of-bound reads/writes, which may trigger errors or warnings.
     """
-    x = torch.randn(128)
-    y = torch.randn(128)
+    x = torch.randn(128, device='cuda')
+    y = torch.randn(128, device='cuda')
     out = torch.empty_like(x)
 
     # The kernel launch uses n_elements=256, exceeding the valid tensor size.

tests/test_print_traceback.py

@@ -24,7 +24,7 @@ def kernel_A(ptr, n):
     tl.store(ptr + pid, val)
 
 def test_print_nested_functions():
-    x = torch.arange(4, dtype=torch.float32)
+    x = torch.arange(4, device='cuda', dtype=torch.float32)
     print("Input:", x)
 
     # We'll launch a grid bigger than x.numel() to force a out-of-bounds error