nicolas
/
ik_llama_opt
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

fix(server): strip executed Bailing blocks whenever calls exist 

The generic template-driven PEG can return structured calls while
leaving the block text in reasoning_content. The rescue-only strip
missed those turns, so raw XML kept reaching client Reasoning panels.
Apply the terminator strip on any final turn carrying tool calls,
regardless of which layer parsed them. Reasoning with no calls is
never touched.
This commit is contained in:
Marvin 2026-09-07 22:07:20 -03:00
parent 253cf76f59
commit f9ffb6fb2d
1 changed files with 11 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
examples/server/server-context.cpp
View File

@ -756,6 +756,17 @@ const common_chat_msg& server_slot::update_chat_msg(bool is_partial, std::vector
bailing::remove_executed_blocks(new_msg.reasoning_content);
}
}
// Same terminator semantic for calls the PEG layer parsed itself: the
// generic template-driven parser can return structured calls while
// leaving the block text in reasoning_content. Strip executed blocks
// whenever the turn carries calls, no matter which layer parsed them.
// Reasoning with no calls is never touched. Revert with the block above.
if (!is_partial && !new_msg.tool_calls.empty()
&& bailing::has_complete_block(new_msg.reasoning_content)) {
new_msg.reasoning_content =
bailing::remove_executed_blocks(new_msg.reasoning_content);
LLAMA_LOG_WARN("Bailing strip: removed executed call block(s) from reasoning_content\n");
}
//new_msg.ensure_tool_call_ids_set(generated_tool_call_ids, gen_tool_call_id);
new_msg.set_tool_call_ids(generated_tool_call_ids, gen_tool_call_id);
chat_msg = new_msg;