Skip to content

server : guard chat-template thinking probe against apply-time jinja errors #24093

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
pwilkin merged 4 commits into from
Jul 9, 2026
Merged

server : guard chat-template thinking probe against apply-time jinja errors #24093

Changes from 3 commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
c8ebb07
server : move chat-template thinking probe inside the init try/catch
palios-taey Jun 3, 2026
063aff5
Merge branch 'master' into fix/chat-template-thinking-guard
palios-taey Jun 25, 2026
d36fc59
Merge branch 'master' into fix/chat-template-thinking-guard
palios-taey Jun 26, 2026
dc30a3a
Merge upstream master into thinking guard
palios-taey Jun 28, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 7 additions & 7 deletions tools/server/server-context.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1496,27 +1496,27 @@ struct server_context_impl {
// populate chat template params
{
common_chat_templates_ptr chat_templates;
bool enable_thinking = false;

try {
chat_templates = common_chat_templates_init(model_tgt, params_base.chat_template);

LOG_INF("%s: chat template, example_format: '%s'\n", __func__,
common_chat_format_example(chat_templates.get(), params_base.use_jinja, params_base.default_template_kwargs).c_str());

// thinking is enabled if:
// 1. It's not explicitly disabled via --reasoning off
// 2. The chat template supports it
const bool template_supports_thinking = params_base.use_jinja && common_chat_templates_support_enable_thinking(chat_templates.get());
enable_thinking = params_base.enable_reasoning != 0 && template_supports_thinking;
SRV_INF("%s: chat template, thinking = %d\n", __func__, enable_thinking);
} catch (const std::exception & e) {
SRV_ERR("%s: chat template parsing error: %s\n", __func__, e.what());
SRV_ERR("%s: please consider disabling jinja via --no-jinja, or use a custom chat template via --chat-template\n", __func__);
SRV_ERR("%s: for example: --no-jinja --chat-template chatml\n", __func__);
return false;
}

// thinking is enabled if:
// 1. It's not explicitly disabled via --reasoning off
// 2. The chat template supports it
const bool template_supports_thinking = params_base.use_jinja && common_chat_templates_support_enable_thinking(chat_templates.get());
const bool enable_thinking = params_base.enable_reasoning != 0 && template_supports_thinking;
SRV_INF("%s: chat template, thinking = %d\n", __func__, enable_thinking);

// IMPORTANT: chat_params is reused across sleeping / resuming states,
// never store llama_context/llama_model pointers in chat_params,
// as they may be invalidated after sleeping
Expand Down