Use inference APIs for executing Llama Guard (#121)

We should use Inference APIs to execute Llama Guard instead of directly needing to use HuggingFace modeling related code. The actual inference consideration is handled by Inference.

llama_stack/providers/utils/inference/__init__.py

@@ -3,3 +3,31 @@
 #
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
+
+from typing import List
+
+from llama_models.datatypes import *  # noqa: F403
+from llama_models.sku_list import all_registered_models
+
+
+def is_supported_safety_model(model: Model) -> bool:
+    if model.quantization_format != CheckpointQuantizationFormat.bf16:
+        return False
+
+    model_id = model.core_model_id
+    return model_id in [
+        CoreModelId.llama_guard_3_8b,
+        CoreModelId.llama_guard_3_1b,
+        CoreModelId.llama_guard_3_11b_vision,
+    ]
+
+
+def supported_inference_models() -> List[str]:
+    return [
+        m.descriptor()
+        for m in all_registered_models()
+        if (
+            m.model_family in {ModelFamily.llama3_1, ModelFamily.llama3_2}
+            or is_supported_safety_model(m)
+        )
+    ]

llama_stack/providers/utils/inference/augment_messages.py

@@ -16,6 +16,8 @@ from llama_models.llama3.prompt_templates import (
 )
 from llama_models.sku_list import resolve_model
 
+from llama_stack.providers.utils.inference import supported_inference_models
+
 
 def augment_messages_for_tools(request: ChatCompletionRequest) -> List[Message]:
     """Reads chat completion request and augments the messages to handle tools.
@@ -27,8 +29,8 @@ def augment_messages_for_tools(request: ChatCompletionRequest) -> List[Message]:
         cprint(f"Could not resolve model {request.model}", color="red")
         return request.messages
 
-    if model.model_family not in [ModelFamily.llama3_1, ModelFamily.llama3_2]:
-        cprint(f"Model family {model.model_family} not llama 3_1 or 3_2", color="red")
+    if model.descriptor() not in supported_inference_models():
+        cprint(f"Unsupported inference model? {model.descriptor()}", color="red")
         return request.messages
 
     if model.model_family == ModelFamily.llama3_1 or (