[mac-ai] Enable RamaLama CPU nightly testing (#823)

Needs containers/ramalama#1909 to work.

<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->

## Summary by CodeRabbit

* **New Features**
* Added a Ramalama “no‑GPU” platform option, available in presets and
platform flavors for easier selection on systems without GPU access.

* **Bug Fixes**
* Non‑GPU runs now use a safe “none” device configuration for improved
stability.
* Automatically disables GPU when the selected inference server flavor
is set to no‑GPU, preventing unintended GPU usage.

<!-- end of auto-generated comment: release notes by coderabbit.ai -->

Author: kpouget

projects/mac_ai/testing/config.yaml

@@ -66,6 +66,7 @@ ci_presets:
     test.model.name: ollama://llama3.2
     test.platform:
     - podman/ramalama
+    - podman/ramalama/no-gpu
     - macos/llama_cpp/vulkan
     # - podman/llama_cpp/vulkan
     - macos/llama_cpp/metal
@@ -75,7 +76,10 @@ ci_presets:
     prepare.ramalama.repo.version: main
     prepare.ramalama.build_image.enabled: true
 
-    test.platform: [podman/ramalama]
+    test.platform:
+    - podman/ramalama/no-gpu
+    - podman/ramalama
+
     test.model.name: ollama://llama3.2
     matbench.lts.generate: true
     test.inference_server.benchmark.enabled: true
@@ -628,4 +632,4 @@ __platform_check:
 
   flavors:
     llama_cpp: [metal, vulkan, upstream_bin, remoting]
-    ramalama: [remoting, null]
+    ramalama: [remoting, no-gpu, null]

projects/mac_ai/testing/ramalama.py

@@ -262,7 +262,7 @@ def _run_from_toolbox(ramalama_cmd, base_work_dir, platform, ramalama_path, mode
 
     want_gpu = platform.want_gpu
     device = config.project.get_config("prepare.podman.container.device") \
-        if want_gpu else "/dev/null"
+        if want_gpu else "none"
 
     if config.project.get_config("prepare.ramalama.build_image.enabled"):
         image_name = config.project.get_config("prepare.ramalama.build_image.name")

projects/mac_ai/testing/utils.py

@@ -61,10 +61,13 @@ def parse_platform(platform_str):
         if platform.inference_server_flavor not in inference_server_has_flavors:
             raise ValueError(f"Invalid platform inference server flavor ({platform.inference_server_flavor}) in {platform_str}. Expected one of {inference_server_has_flavors}")
 
+
     no_gpu_option_name = config.project.get_config("__platform_check.options.no_gpu", print=False)
     if no_gpu_option_name in platform_parts:
         platform.want_gpu = False
         platform_parts.remove(no_gpu_option_name)
+    elif platform.inference_server_flavor == no_gpu_option_name:
+        platform.want_gpu = False
     else:
         platform.want_gpu = True