updated test to mark as potential failure

src/c++/perf_analyzer/genai-perf/tests/test_llm_inputs.py

@@ -189,6 +189,119 @@ def test_llm_inputs_with_defaults(self, default_configured_url):
         dataset_json = LlmInputs._convert_input_url_dataset_to_generic_json(
             dataset=dataset
         )
+        pa_json = LlmInputs._convert_generic_json_to_output_format(
+            output_format=OutputFormat.OPENAI_CHAT_COMPLETIONS,
+            generic_dataset=dataset_json,
+            add_model_name=False,
+            add_stream=False,
+            extra_inputs={},
+            output_tokens_mean=LlmInputs.DEFAULT_OUTPUT_TOKENS_MEAN,
+            output_tokens_stddev=LlmInputs.DEFAULT_OUTPUT_TOKENS_STDDEV,
+            output_tokens_deterministic=False,
+        )
+
+        assert pa_json is not None
+        assert len(pa_json["data"]) == LlmInputs.DEFAULT_LENGTH
+
+    @pytest.mark.xfail(
+        reason="Download from huggingface may fail due to server issues",
+        raises=GenAIPerfException,
+    )
+    def test_create_openai_llm_inputs_cnn_dailymail(self):
+        """
+        Test CNN_DAILYMAIL can be accessed
+        """
+        pa_json = LlmInputs.create_llm_inputs(
+            input_type=PromptSource.DATASET,
+            dataset_name=CNN_DAILY_MAIL,
+            output_format=OutputFormat.OPENAI_CHAT_COMPLETIONS,
+        )
+
+        os.remove(DEFAULT_INPUT_DATA_JSON)
+
+        assert pa_json is not None
+        assert len(pa_json["data"]) == LlmInputs.DEFAULT_LENGTH
+
+    def test_write_to_file(self):
+        """
+        Test that write to file is working correctly
+        """
+        pa_json = LlmInputs.create_llm_inputs(
+            input_type=PromptSource.DATASET,
+            dataset_name=OPEN_ORCA,
+            output_format=OutputFormat.OPENAI_CHAT_COMPLETIONS,
+            model_name="open_orca",
+            add_model_name=True,
+            add_stream=True,
+        )
+        try:
+            with open(DEFAULT_INPUT_DATA_JSON, "r") as f:
+                json_str = f.read()
+        finally:
+            os.remove(DEFAULT_INPUT_DATA_JSON)
+
+        assert pa_json == json.loads(json_str)
+
+    def test_create_openai_to_vllm(self):
+        """
+        Test conversion of openai to vllm
+        """
+        pa_json = LlmInputs.create_llm_inputs(
+            input_type=PromptSource.DATASET,
+            output_format=OutputFormat.VLLM,
+            dataset_name=OPEN_ORCA,
+            add_model_name=False,
+            add_stream=True,
+        )
+
+        os.remove(DEFAULT_INPUT_DATA_JSON)
+
+        assert pa_json is not None
+        assert len(pa_json["data"]) == LlmInputs.DEFAULT_LENGTH
+
+    def test_create_openai_to_completions(self):
+        """
+        Test conversion of openai to completions
+        """
+        pa_json = LlmInputs.create_llm_inputs(
+            input_type=PromptSource.DATASET,
+            output_format=OutputFormat.OPENAI_COMPLETIONS,
+            dataset_name=OPEN_ORCA,
+            add_model_name=False,
+            add_stream=True,
+        )
+
+        os.remove(DEFAULT_INPUT_DATA_JSON)
+
+        assert pa_json is not None
+        assert len(pa_json["data"]) == LlmInputs.DEFAULT_LENGTH
+        # NIM legacy completion endpoint only supports string and not
+        # array of strings. Verify that the prompt is of type string
+        # not list
+        assert isinstance(pa_json["data"][0]["payload"][0]["prompt"], str)
+
+    def test_create_openai_to_trtllm(self):
+        """
+        Test conversion of openai to trtllm
+        """
+        pa_json = LlmInputs.create_llm_inputs(
+            input_type=PromptSource.DATASET,
+            output_format=OutputFormat.TENSORRTLLM,
+            dataset_name=OPEN_ORCA,
+            add_model_name=False,
+            add_stream=True,
+        )
+
+        os.remove(DEFAULT_INPUT_DATA_JSON)
+
+        assert pa_json is not None
+        assert len(pa_json["data"]) == LlmInputs.DEFAULT_LENGTH
+
+    def test_random_synthetic_no_stddev(self, default_tokenizer):
+        """
+        Test that we can produce an exact number of random synthetic tokens
+        """
+        random.seed(1)
 
         assert dataset_json is not None
         assert len(dataset_json["rows"]) == TEST_LENGTH