[TP-1781] Add test for MiniCPM-o-2_6-language model

tests/common.py

@@ -65,6 +65,23 @@ def metadata(pat: bool = False) -> Tuple[Tuple[str, str], Tuple[str, str]]:
     )
 
 
+def grpc_channel(func):
+    """
+    A decorator that runs the test using the gRPC channel.
+    :param func: The test function.
+    :return: A function wrapper.
+    """
+
+    def func_wrapper():
+        if os.getenv("CLARIFAI_GRPC_INSECURE", "False").lower() in ("true", "1", "t"):
+            channel = ClarifaiChannel.get_insecure_grpc_channel(port=443)
+        else:
+            channel = ClarifaiChannel.get_grpc_channel()
+        func(channel)
+
+    return func_wrapper
+
+
 def both_channels(func):
     """
     A decorator that runs the test first using the gRPC channel and then using the JSON channel.
@@ -73,9 +90,10 @@ def both_channels(func):
     """
 
     def func_wrapper():
-        channel = ClarifaiChannel.get_grpc_channel()
         if os.getenv("CLARIFAI_GRPC_INSECURE", "False").lower() in ("true", "1", "t"):
             channel = ClarifaiChannel.get_insecure_grpc_channel(port=443)
+        else:
+            channel = ClarifaiChannel.get_grpc_channel()
         func(channel)
 
         channel = ClarifaiChannel.get_json_channel()
@@ -335,6 +353,25 @@ def post_model_outputs_and_maybe_allow_retries(
     return response
 
 
+def _generate_model_outputs(
+    stub: service_pb2_grpc.V2Stub,
+    request: service_pb2.PostModelOutputsRequest,
+    metadata: Tuple,
+):
+    is_model_loaded = False
+    for i in range(1, MAX_PREDICT_ATTEMPTS + 1):
+        response_iterator = stub.GenerateModelOutputs(request, metadata=metadata)
+        for response in response_iterator:
+            if not is_model_loaded and response.status.code == status_code_pb2.MODEL_LOADING:
+                print(f"Model {request.model_id} is still loading...")
+                time.sleep(15)
+                break
+            is_model_loaded = True
+            yield response
+        if is_model_loaded:
+            break
+
+
 async def async_post_model_outputs_and_maybe_allow_retries(
     stub: service_pb2_grpc.V2Stub,
     request: service_pb2.PostModelOutputsRequest,

tests/public_models/public_test_helper.py

@@ -31,6 +31,7 @@
 TEXT_SENTIMENT_MODEL_ID = "multilingual-uncased-sentiment"  # bert-based
 TEXT_MULTILINGUAL_MODERATION_MODEL_ID = "bdcedc0f8da58c396b7df12f634ef923"
 
+TEXT_MINI_CPM_MODEL_ID = "MiniCPM-o-2_6-language"
 
 TRANSLATION_TEST_DATA = {
     "ROMANCE": "No me apetece nada estudiar esta noche",
@@ -212,6 +213,10 @@
     ),
 ]
 
+TEXT_LLM_MODEL_TITLE_IDS_TUPLE = [
+    ("multimodal large language model", TEXT_MINI_CPM_MODEL_ID, "miniCPM", "openbmb")
+]
+
 # title, model_id, text, app, user
 TEXT_FB_TRANSLATION_MODEL_TITLE_ID_DATA_TUPLE = []
 TEXT_HELSINKI_TRANSLATION_MODEL_TITLE_ID_DATA_TUPLE = []

tests/public_models/test_public_models_predicts.py

@@ -19,10 +19,12 @@
     GENERAL_MODEL_ID,
     MAIN_APP_ID,
     MAIN_APP_USER_ID,
+    _generate_model_outputs,
     async_post_model_outputs_and_maybe_allow_retries,
     async_raise_on_failure,
     asyncio_channel,
     both_channels,
+    grpc_channel,
     metadata,
     post_model_outputs_and_maybe_allow_retries,
     raise_on_failure,
@@ -35,6 +37,7 @@
     MULTIMODAL_MODEL_TITLE_AND_IDS,
     TEXT_FB_TRANSLATION_MODEL_TITLE_ID_DATA_TUPLE,
     TEXT_HELSINKI_TRANSLATION_MODEL_TITLE_ID_DATA_TUPLE,
+    TEXT_LLM_MODEL_TITLE_IDS_TUPLE,
     TEXT_MODEL_TITLE_IDS_TUPLE,
     TRANSLATION_TEST_DATA,
 )
@@ -120,6 +123,60 @@ def test_text_predict_on_public_models(channel):
         )
 
 
+@grpc_channel
+def test_text_predict_on_public_llm_models(channel):
+    stub = service_pb2_grpc.V2Stub(channel)
+
+    for title, model_id, app_id, user_id in TEXT_LLM_MODEL_TITLE_IDS_TUPLE:
+        request = service_pb2.PostModelOutputsRequest(
+            user_app_id=resources_pb2.UserAppIDSet(user_id=user_id, app_id=app_id),
+            model_id=model_id,
+            inputs=[
+                resources_pb2.Input(
+                    data=resources_pb2.Data(
+                        parts=[
+                            resources_pb2.Part(
+                                id="prompt",
+                                data=resources_pb2.Data(
+                                    string_value=TRANSLATION_TEST_DATA["EN"],
+                                ),
+                            ),
+                            resources_pb2.Part(
+                                id="max_tokens",
+                                data=resources_pb2.Data(
+                                    int_value=10,
+                                ),
+                            ),
+                            resources_pb2.Part(
+                                id="temperature",
+                                data=resources_pb2.Data(
+                                    float_value=0.7,
+                                ),
+                            ),
+                            resources_pb2.Part(
+                                id="top_p",
+                                data=resources_pb2.Data(
+                                    float_value=0.95,
+                                ),
+                            ),
+                        ]
+                    )
+                )
+            ],
+        )
+        response_iterator = _generate_model_outputs(stub, request, metadata(pat=True))
+
+        responses_count = 0
+        for response in response_iterator:
+            responses_count += 1
+            raise_on_failure(
+                response,
+                custom_message=f"Text predict failed for the {title} model (ID: {model_id}).",
+            )
+
+        assert responses_count > 0
+
+
 @asyncio_channel
 async def test_text_predict_on_public_models_async(channel):
     """Test non translation text/nlp models.