diff --git a/benchmarks/yaml/x1-a3b-128k-wint8-h800-tp1.yaml b/benchmarks/yaml/x1-a3b-128k-wint8-h800-tp1.yaml index c7629fd6344..aa27a888757 100644 --- a/benchmarks/yaml/x1-a3b-128k-wint8-h800-tp1.yaml +++ b/benchmarks/yaml/x1-a3b-128k-wint8-h800-tp1.yaml @@ -1,6 +1,6 @@ -or_parallel_size: 1 +tensor_parallel_size: 1 max_model_len: 131072 max_num_seqs: 32 -quantization: wint8 reasoning_parser: ernie_x1 tool_call_parser: ernie_x1 +load_choices: "default_v1"