ClaudeBot/llama.cpp
0
0
Fork 1
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-09-17 20:31:47 +02:00
Code Issues Packages Projects Releases Wiki Activity

chat : force \n</think> on reasoning budget end for qwen3-coder (#28869)

Browse Source
This commit is contained in:
Aldehir Rojas
2026-09-16 08:47:28 +02:00
committed by GitHub
parent d4365d9554
commit 0bec16e388
1 changed files with 2 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
common/parsers/qwen3-coder.cpp
+2 -1
View File
@@ -23,8 +23,9 @@ common_chat_params common_chat_params_init_qwen3_coder(const common_chat_templat
if (supports_reasoning) {
data.thinking_start_tag = "<think>";
// Support both </think> and <tool_call> as reasoning end sequences.
// The newline variant comes first so it is included in the forced message
// <function= is omitted, as it is a workaround for Qwen3-Coder which is not a thinking model
data.thinking_end_tags = { "</think>", "<tool_call>" };
data.thinking_end_tags = { "\n</think>", "</think>", "<tool_call>" };
data.preserved_tokens.insert(data.preserved_tokens.end(), { "<think>", "</think>" });
}