Commit 1052f7e1d for llama.cpp

commit 1052f7e1d62611d328ccc65d38c77cd194ee59fc
Author: Tarek Dakhran <tarek@liquid.ai>
Date:   Sat Oct 31 04:38:04 2026 +0100

    chat : strip the think block in the qwen3-coder with reasoning_format none (#31380)

    * chat : accept the think block in the qwen3-coder response format with reasoning_format none

    Assisted-by: Claude Opus 5.5

    * chat : define both qwen3-coder think block variants up front

diff --git a/common/parsers/qwen3-coder.cpp b/common/parsers/qwen3-coder.cpp
index 7cb9cce90..a68a6835e 100644
--- a/common/parsers/qwen3-coder.cpp
+++ b/common/parsers/qwen3-coder.cpp
@@ -74,16 +74,24 @@ common_chat_params common_chat_params_init_qwen3_coder(const common_chat_templat
     data.parser = build_chat_peg_parser([&](common_chat_peg_builder & p) {
         auto generation_prompt = p.literal(GEN_PREFIX);

+        // The template strips the newline before </think>, so leave it out of the reasoning
+        auto think_body  = p.until_one_of({ "\n</think>", "</think>", "<tool_call>" });
+        auto think_close = p.literal("\n</think>") | p.literal("</think>") | p.peek(p.literal("<tool_call>"));
+
+        auto reasoning_extracted = p.optional("<think>" + p.space() + p.reasoning(think_body) + think_close);
+        auto reasoning_skipped   = p.optional("<think>" + p.space() + think_body + think_close);
+
         auto reasoning = p.eps();
         if (supports_reasoning && extract_reasoning) {
-            // The template strips the newline before </think>, so leave it out of the reasoning
-            reasoning = p.optional("<think>" + p.space() +
-                                   p.reasoning(p.until_one_of({ "\n</think>", "</think>", "<tool_call>" })) +
-                                   (p.literal("\n</think>") | p.literal("</think>") | p.peek(p.literal("<tool_call>"))));
+            reasoning = reasoning_extracted;
         }

         // Response format parser
         if (has_response_format) {
+            // The answer must be bare JSON, so the think block is consumed even when it is not extracted
+            if (supports_reasoning && !extract_reasoning) {
+                reasoning = reasoning_skipped;
+            }
             return generation_prompt + (reasoning << p.content(p.schema(p.json(), "response-format", inputs.json_schema)));
         }

diff --git a/tests/test-chat.cpp b/tests/test-chat.cpp
index 0ab8e65ab..fa4525389 100644
--- a/tests/test-chat.cpp
+++ b/tests/test-chat.cpp
@@ -5629,6 +5629,31 @@ static void test_template_output_peg_parsers(bool detailed_debug) {
             .expect(message_assist_call_thoughts)
             .run();

+        const std::string answer_schema = R"({"type":"object","properties":{"answer":{"type":"integer"}},"required":["answer"]})";
+
+        tst.test("I'm\nthinking</think>{\"answer\":42}")
+            .enable_thinking(true)
+            .reasoning_format(COMMON_REASONING_FORMAT_AUTO)
+            .json_schema(answer_schema)
+            .expect_reasoning("I'm\nthinking")
+            .expect_content(R"({"answer":42})")
+            .run();
+
+        // the reasoning block of the generation prompt must still be accepted when it is not extracted
+        tst.test("{\"answer\":42}")
+            .enable_thinking(false)
+            .reasoning_format(COMMON_REASONING_FORMAT_NONE)
+            .json_schema(answer_schema)
+            .expect_content(R"({"answer":42})")
+            .run();
+
+        tst.test("I'm\nthinking</think>{\"answer\":42}")
+            .enable_thinking(true)
+            .reasoning_format(COMMON_REASONING_FORMAT_NONE)
+            .json_schema(answer_schema)
+            .expect_content(R"({"answer":42})")
+            .run();
+
         // a deferred tool is left out of <tools>, a tool_reference renders it, and its schema types the call
         const auto deferred_tools = common_chat_tools_parse_oaicompat(json::parse(R"([
             {"type": "function", "function": {"name": "tool_search", "description": "Search for tools",