sampling : fix off-by-one in tail-free sampling

ggml-ci

common/arg.cpp

@@ -963,7 +963,7 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
         }
     ).set_sparam());
     add_opt(llama_arg(
-        {"--tfs"}, "N",
+        {"--tfs", "--tfs-z"}, "Z",
         format("tail free sampling, parameter z (default: %.1f, 1.0 = disabled)", (double)params.sparams.tfs_z),
         [](gpt_params & params, const std::string & value) {
             params.sparams.tfs_z = std::stof(value);