common, server : enable preserve_reasoning kwarg by default, log its effective state (#28174)

* common, server : enable preserve_reasoning kwarg by default, log its effective state

If the preserve_reasoning chat template kwarg is not specified explicitly
via --reasoning-preserve / --no-reasoning-preserve, it is enabled by
default after argument processing. The server logs the effective state of
the kwarg, warns that it is enabled by default when the template supports
it, and only warns "has no effect" when it was enabled explicitly on a
template that does not support it. Setting the kwarg via
--chat-template-kwargs is deprecated.

Assisted-by: pi:llama.cpp/Qwen3.8-27B

* cont : update comment

Co-authored-by: Xuan-Son Nguyen <son@huggingface.co>

---------

Co-authored-by: Xuan-Son Nguyen <son@huggingface.co>
This commit is contained in:
Georgi Gerganov
2026-09-02 19:19:54 +03:00
committed by GitHub
co-authored by Xuan-Son Nguyen
parent 7798007a29
commit e750b887a8
3 changed files with 26 additions and 4 deletions
+11 -1
View File
@@ -960,6 +960,11 @@ static bool common_params_parse_ex(int argc, char ** argv, common_params_context
));
}
// if the preserve_reasoning kwarg was not specified explicitly, enable it by default
if (!params.default_template_kwargs.count("preserve_reasoning")) {
params.default_template_kwargs["preserve_reasoning"] = "true";
}
return true;
}
@@ -3553,6 +3558,10 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
LOG_WRN("Setting 'enable_thinking' via --chat-template-kwargs is deprecated. "
"Use --reasoning on / --reasoning off instead.\n");
}
if (item.key() == "preserve_reasoning") {
LOG_WRN("Setting 'preserve_reasoning' via --chat-template-kwargs is deprecated. "
"Use --reasoning-preserve / --no-reasoning-preserve instead.\n");
}
params.default_template_kwargs[item.key()] = item.value().dump();
}
}
@@ -3743,7 +3752,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
add_opt(common_arg(
{"--reasoning-preserve"},
{"--no-reasoning-preserve"},
"preserve reasoning trace in the full history, not just the last assistant message (default: template default)\n"
"preserve reasoning trace in the full history, not just the last assistant message (default: enabled)\n"
"compatible with certain templates having 'supports_preserve_reasoning' capability\n"
"example: https://docs.z.ai/guides/capabilities/thinking-mode#preserved-thinking",
[](common_params & params, bool value) {
@@ -3752,6 +3761,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
} else {
params.default_template_kwargs["preserve_reasoning"] = "false";
}
params.preserve_reasoning_specified = true;
}
).set_examples({LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_COMPLETION, LLAMA_EXAMPLE_CLI}).set_env("LLAMA_ARG_REASONING_PRESERVE"));
add_opt(common_arg(