From 1110e3d89035e7df56422cd5c4c715a90e515f3f Mon Sep 17 00:00:00 2001 From: Izzy Putterman <iputterman@nvidia.com> Date: Mon, 13 May 2024 17:22:20 -0700 Subject: [PATCH] Bug fixes --- .../genai-perf/genai_perf/llm_inputs/llm_inputs.py | 4 ++-- src/c++/perf_analyzer/genai-perf/genai_perf/llm_metrics.py | 2 +- 2 files changed, 3 insertions(+), 3 deletions(-) diff --git a/src/c++/perf_analyzer/genai-perf/genai_perf/llm_inputs/llm_inputs.py b/src/c++/perf_analyzer/genai-perf/genai_perf/llm_inputs/llm_inputs.py index c3878b9c3..1431e9a65 100644 --- a/src/c++/perf_analyzer/genai-perf/genai_perf/llm_inputs/llm_inputs.py +++ b/src/c++/perf_analyzer/genai-perf/genai_perf/llm_inputs/llm_inputs.py @@ -579,7 +579,7 @@ def _convert_generic_json_to_trtllm_backend_format( text_input_headers, ) = cls._determine_json_feature_roles(dataset_json) - pa_json = cls._populate_trtllm_output_json( + pa_json = cls._populate_trtllm_backend_output_json( dataset_json, system_role_headers, user_role_headers, @@ -819,7 +819,7 @@ def _populate_trtllm_output_json( return pa_json - @classmethod + @classmethod def _populate_trtllm_backend_output_json( cls, dataset_json: Dict, diff --git a/src/c++/perf_analyzer/genai-perf/genai_perf/llm_metrics.py b/src/c++/perf_analyzer/genai-perf/genai_perf/llm_metrics.py index 0928c75a9..4856a5d82 100755 --- a/src/c++/perf_analyzer/genai-perf/genai_perf/llm_metrics.py +++ b/src/c++/perf_analyzer/genai-perf/genai_perf/llm_metrics.py @@ -652,7 +652,7 @@ def _tokenize_openai_request_input(self, req_inputs: dict) -> List[int]: def _tokenize_response_outputs(self, res_outputs: dict) -> List[List[int]]: """Deserialize the response output and return tokenized outputs.""" if self._service_kind == "triton" and self._response_format == ResponseFormat.TENSORRTLLM_BACKEND: - return self._tokenize_trtllm_response_output(req_inputs) + return self._tokenize_trtllm_response_output(res_outputs) elif self._service_kind == "triton": return self._tokenize_triton_response_output(res_outputs) elif self._service_kind == "openai":