add cuda-tests

Signed-off-by: sallyom <somalley@redhat.com>
containers · Apr 12, 2024 · 9f93d55 · 9f93d55
1 parent 3f081e8
commit 9f93d55
Show file tree

Hide file tree

Showing 4 changed files with 15 additions and 10 deletions.
diff --git a/.github/workflows/model_servers.yaml b/.github/workflows/model_servers.yaml
@@ -25,29 +25,33 @@ jobs:
       matrix:
         include:
           - image_name: llamacpp_python
+            cuda_image_name: llamacpp_python_cuda
+            base_image_name: llamacpp_python
             model: mistral
             flavor: base
             directory: llamacpp_python
             platforms: linux/amd64,linux/arm64
-            test_cmd: test
           #- image_name: llamacpp_python_vulkan
+          #  cuda_image_name: llamacpp_python_cuda
+          #  base_image_name: llamacpp_python
           #  model: mistral
           #  flavor: vulkan
           #  directory: llamacpp_python
           #  platforms: linux/arm64
-          #  test_cmd: test
           - image_name: llamacpp_python_cuda
+            cuda_image_name: llamacpp_python_cuda
+            base_image_name: llamacpp_python
             model: mistral
             flavor: cuda
             directory: llamacpp_python
             platforms: linux/amd64
-            test_cmd: run-cuda
           - image_name: whispercpp
+            cuda_image_name: llamacpp_python_cuda
+            base_image_name: llamacpp_python
             model: whisper-small
             flavor: base
             directory: whispercpp
             platforms: linux/amd64,linux/arm64
-            test_cmd: test
     runs-on: ubuntu-latest
     permissions:
       contents: read
@@ -95,9 +99,11 @@ jobs:
 
       - name: Run tests
         working-directory: ./model_servers/${{ matrix.directory }}/
-        run: make ${{ matrix.test_cmd }}
+        run: make test
         env:
           IMAGE_NAME: ${{ matrix.image_name }}
+          BASE_IMAGE_NAME: ${{ matrix.base_image_name }}
+          CUDA_IMAGE_NAME: ${{ matrix.cuda_image_name }}
 
       - name: Login to Container Registry
         if: github.event_name == 'push' && github.ref == 'refs/heads/main'

diff --git a/model_servers/llamacpp_python/Makefile b/model_servers/llamacpp_python/Makefile
@@ -68,11 +68,10 @@ run:
 	cd ../../models && \
 	podman run -it -d -p $(PORT):$(PORT) -v ./$(SELECTED_MODEL_NAME):$(MODELS_PATH)/model.gguf:$(BIND_MOUNT_OPTIONS) -e MODEL_PATH=$(MODELS_PATH)/model.gguf -e HOST=0.0.0.0 -e PORT=$(PORT) --net=host $(IMAGE)
 
-# TODO: set to fail if container isn't running (set || true to enable tests temporarily)
 .PHONY: run-cuda
 run-cuda:
 	cd ../../models && \
-	podman run -it -d -p $(PORT):$(PORT) -v ./$(SELECTED_MODEL_NAME):$(MODELS_PATH)/model.gguf:$(BIND_MOUNT_OPTIONS) -e MODEL_PATH=$(MODELS_PATH)/model.gguf -e HOST=0.0.0.0 -e PORT=$(PORT) --net=host --device nvidia.com/gpu=all $(IMAGE) || true
+	podman run -it -d -p $(PORT):$(PORT) -v ./$(SELECTED_MODEL_NAME):$(MODELS_PATH)/model.gguf:$(BIND_MOUNT_OPTIONS) -e MODEL_PATH=$(MODELS_PATH)/model.gguf -e HOST=0.0.0.0 -e PORT=$(PORT) --net=host --device nvidia.com/gpu=all $(IMAGE)
 
 .PHONY: test
 test:

diff --git a/model_servers/llamacpp_python/tests/conftest.py b/model_servers/llamacpp_python/tests/conftest.py
@@ -2,7 +2,7 @@
 import os
 
 CUDA_MS = pytest_container.Container(
-        url=f"containers-storage:{os.environ['REGISTRY']}/containers/{os.environ['IMAGE_NAME']}",
+        url=f"containers-storage:{os.environ['REGISTRY']}/containers/{os.environ['CUDA_IMAGE_NAME']}",
         volume_mounts=[
             pytest_container.container.BindMount(
                 container_path="/locallm/models/model.gguf",
@@ -25,7 +25,7 @@
     )
 
 MS = pytest_container.Container(
-        url=f"containers-storage:{os.environ['REGISTRY']}/containers/{os.environ['IMAGE_NAME']}",
+        url=f"containers-storage:{os.environ['REGISTRY']}/containers/{os.environ['BASE_IMAGE_NAME']}",
         volume_mounts=[
             pytest_container.container.BindMount(
                 container_path="/locallm/models/model.gguf",

diff --git a/model_servers/llamacpp_python/tests/test_alive.py b/model_servers/llamacpp_python/tests/test_alive.py
@@ -1,10 +1,10 @@
 import pytest_container
 from .conftest import MS
+from .conftest import CUDA_MS
 import tenacity
 
 CONTAINER_IMAGES = [MS, CUDA_MS]
 
-
 def test_etc_os_release_present(auto_container: pytest_container.container.ContainerData):
     assert auto_container.connection.file("/etc/os-release").exists