Fix

llama_stack/providers/impls/meta_reference/inference/inference.py

@@ -95,7 +95,7 @@ class MetaReferenceInferenceImpl(Inference, ModelsProtocolPrivate):
         def impl():
             stop_reason = None
 
-            for token_result in self.generator.chat_completion(request):
+            for token_result in self.generator.completion(request):
                 if token_result.text == "<|eot_id|>":
                     stop_reason = StopReason.end_of_turn
                     text = ""

llama_stack/providers/tests/inference/test_inference.py

@@ -159,7 +159,10 @@ async def test_completion(inference_settings):
         )
     ]
 
-    print(chunks)
+    assert all(isinstance(chunk, CompletionResponseStreamChunk) for chunk in chunks)
+    assert len(chunks) == 51
+    last = chunks[-1]
+    assert last.stop_reason == StopReason.out_of_tokens
 
 
 @pytest.mark.asyncio