Modal tests

.github/workflows/testing.yml

@@ -15,7 +15,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: astral-sh/setup-uv@v4
       - run: uv sync --extra dev
-      - run: uv run pytest --cov-report term --cov-report html --cov-report xml --cov=src/libkernelbot unit-tests -v
+      - run: uv run pytest --cov-report term --cov-report html --cov-report xml --cov=src/libkernelbot -m "not integration" tests -v
       - uses: actions/upload-artifact@v4
         with:
           name: coverage
@@ -25,13 +25,22 @@ jobs:
         uses: py-cov-action/python-coverage-comment-action@v3
         with:
           GITHUB_TOKEN: ${{ github.token }}
-
       - name: Store Pull Request comment to be posted
         uses: actions/upload-artifact@v4
         if: steps.coverage_comment.outputs.COMMENT_FILE_WRITTEN == 'true'
         with:
-          # If you use a different name, update COMMENT_ARTIFACT_NAME accordingly
           name: python-coverage-comment-action
-          # If you use a different name, update COMMENT_FILENAME accordingly
           path: python-coverage-comment-action.txt
 
+  integration-tests:
+    runs-on: ubuntu-latest
+    timeout-minutes: 30
+    env:
+      MODAL_TOKEN_ID: ${{ secrets.MODAL_TOKEN_ID }}
+      MODAL_TOKEN_SECRET: ${{ secrets.MODAL_TOKEN_SECRET }}
+    steps:
+      - uses: actions/checkout@v4
+      - uses: astral-sh/setup-uv@v4
+      - run: uv sync --extra dev
+      - run: uv run modal token set --token-id ${MODAL_TOKEN_ID} --token-secret ${MODAL_TOKEN_SECRET}
+      - run: uv run pytest -m integration tests -v

examples/vectoradd_py/submission_cuda_inline.py

@@ -54,7 +54,6 @@
 """
 
 
-
 add_module = load_inline(
     name='add_cuda',
     cpp_sources=add_cpp_source,
@@ -63,62 +62,12 @@
     verbose=True,
 )
 
+
 def add(A, B):
     if not A.is_cuda or not B.is_cuda:
         raise RuntimeError("Both tensors must be on GPU")
     return add_module.add_cuda(A, B)
 
-def custom_kernel(data: input_t) -> output_t:
-    """
-    Custom implementation of vector addition using CUDA inline function.
-    Args:
-        inputs: List of pairs of tensors [A, B] to be added.
-    Returns:
-        List of tensors containing element-wise sums.
-    """
-    A, B = data
 
-    assert A.is_cuda and B.is_cuda, "Input tensors must be on GPU"
-    assert A.shape == B.shape, "Input tensors must have the same shape"
-    assert A.dtype == torch.float16 and B.dtype == torch.float16, "Input tensors must be float16"
-
-    M, N = A.shape
-    C = torch.empty_like(A)
-
-    n_threads = 256
-    n_blocks = (M * N + n_threads - 1) // n_threads
-
-    cuda_source = """
-    extern "C" __global__ void add_kernel(
-        const half* __restrict__ A,
-        const half* __restrict__ B,
-        half* __restrict__ C,
-        const int n_elements
-    ) {
-        const int idx = blockIdx.x * blockDim.x + threadIdx.x;
-        if (idx < n_elements) {
-            C[idx] = __hadd(A[idx], B[idx]);
-        }
-    }
-    """
-
-    module = torch.utils.cpp_extension.load_inline(
-        name=f"add_kernel_{M}_{N}",
-        cpp_sources="",
-        cuda_sources=cuda_source,
-        functions=["add_kernel"],
-        with_cuda=True,
-        extra_cuda_cflags=["-arch=sm_70"],  # Adjust based on your GPU architecture
-    )
-
-    module.add_kernel(
-        cuda_stream=torch.cuda.current_stream(),
-        args=[
-            A.reshape(-1), B.reshape(-1), C.reshape(-1),
-            M * N,
-        ],
-        blocks=n_blocks,
-        threads=n_threads,
-    )
-
-    return C
+def custom_kernel(data: input_t) -> output_t:
+    return add(*data)

pyproject.toml

@@ -52,6 +52,9 @@ exclude_lines = [
 [tool.pytest.ini_options]
 testpaths = ["scripts", "tests"]
 python_files = ["test_*.py", "*_test.py", "ci_test_*.py"]
+markers = [
+    "integration: integration tests that need to interact externally, e.g., with modal/github actions/etc"
+]
 
 [tool.ruff]
 line-length = 120

src/libkernelbot/launchers/modal.py

@@ -32,7 +32,7 @@ async def run_submission(
 
         result = await loop.run_in_executor(
             None,
-            lambda: modal.Function.lookup("discord-bot-runner", func_name).remote(config=config),
+            lambda: modal.Function.from_name("discord-bot-runner", func_name).remote(config=config),
         )
 
         await status.update("✅ Waiting for modal run to finish... Done")

src/runners/modal_runner.py

@@ -33,10 +33,10 @@
         "PyYAML",
     )
     .pip_install(
-        "torch~=2.7",
+        "torch>=2.7.0,<2.8.0",
         "torchvision~=0.22",
-        "torchaudio~=2.7",
-        index_url="https://download.pytorch.org/whl/cu128"
+        "torchaudio>=2.7.0,<2.8.0",
+        index_url="https://download.pytorch.org/whl/cu128",
     )
     # other frameworks
     .pip_install(

unit-tests/conftest.py → tests/conftest.py

@@ -8,14 +8,10 @@
 
 
 @pytest.fixture(scope="module")
-def docker_compose():
-    tgt_path = Path.cwd()
-    if tgt_path.name == "unit-tests":
-        tgt_path = tgt_path.parent
-
+def docker_compose(project_root: Path):
     """Start a test database and run migrations"""
     subprocess.check_call(
-        ["docker", "compose", "-f", "docker-compose.test.yml", "up", "-d"], cwd=tgt_path
+        ["docker", "compose", "-f", "docker-compose.test.yml", "up", "-d"], cwd=project_root
     )
 
     try:
@@ -25,7 +21,7 @@ def docker_compose():
                 ["docker", "compose", "-f", "docker-compose.test.yml", "ps", "-q", "migrate-test"],
                 capture_output=True,
                 text=True,
-                cwd=tgt_path,
+                cwd=project_root,
             )
 
             if not result.stdout.strip():  # Container no longer exists
@@ -37,7 +33,7 @@ def docker_compose():
             ["docker", "compose", "-f", "docker-compose.test.yml", "logs", "migrate-test"],
             capture_output=True,
             text=True,
-            cwd=tgt_path,
+            cwd=project_root,
         )
 
         if "error" in logs.stdout.lower():
@@ -46,7 +42,7 @@ def docker_compose():
         yield
     finally:
         subprocess.run(
-            ["docker", "compose", "-f", "docker-compose.test.yml", "down", "-v"], cwd=tgt_path
+            ["docker", "compose", "-f", "docker-compose.test.yml", "down", "-v"], cwd=project_root
         )
 
 
@@ -122,3 +118,8 @@ def task_directory(tmp_path):
     # Create task.yml
     Path.write_text(tmp_path / "task.yml", TASK_YAML)
     return tmp_path
+
+
+@pytest.fixture(scope="session")
+def project_root():
+    return Path(__file__).parent.parent