diff --git a/python/llm/src/bigdl/llm/cli/llm-cli b/python/llm/src/bigdl/llm/cli/llm-cli index 07562747..a145c09a 100755 --- a/python/llm/src/bigdl/llm/cli/llm-cli +++ b/python/llm/src/bigdl/llm/cli/llm-cli @@ -47,7 +47,11 @@ function starcoder { } function chatglm { - command="$lib_dir/main-chatglm_vnni -t $threads -n $n_predict ${filteredArguments[*]}" + if [[ $(lscpu | grep "amx_int8") ]]; then + command="$lib_dir/main-chatglm_amx -t $threads -n $n_predict ${filteredArguments[*]}" + else + command="$lib_dir/main-chatglm_vnni -t $threads -n $n_predict ${filteredArguments[*]}" + fi echo "$command" eval "$command" }