chat : stop qwen3-coder reasoning and content before template separators (#30557)

* chat : strip trailing whitespace from parsed reasoning and content

Assisted-by: Claude Fable 5.1

* chat : stop qwen3-coder reasoning and content before template separators

Assisted-by: Claude

* tests : remove the Qwen parser case from the chat session tests

Assisted-by: Claude

* common : handle partial prefix until match at end of generation

---------

Co-authored-by: Zeshawn Zahid <zzshawnn@gmail.com>
Co-authored-by: Alde Rojas <hello@alde.dev>

common/chat.cpp

@@ -1803,6 +1803,9 @@ common_chat_msg common_chat_peg_parse(const common_peg_arena &          src_pars
     //LOG_DBG("Parsing PEG input with format %s: %s\n", common_chat_format_name(params.format), effective_input.text.c_str());
 
     common_peg_parse_flags flags = COMMON_PEG_PARSE_FLAG_LENIENT;
+    if (!is_partial) {
+        flags |= COMMON_PEG_PARSE_FLAG_COMPLETE;
+    }
     if (params.debug) {
         flags |= COMMON_PEG_PARSE_FLAG_DEBUG;
     }

common/parsers/qwen3-coder.cpp

@@ -76,9 +76,10 @@ common_chat_params common_chat_params_init_qwen3_coder(const common_chat_templat
 
         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({ "</think>", "<tool_call>" })) +
-                                   (p.literal("</think>") | p.peek(p.literal("<tool_call>"))));
+                                   p.reasoning(p.until_one_of({ "\n</think>", "</think>", "<tool_call>" })) +
+                                   (p.literal("\n</think>") | p.literal("</think>") | p.peek(p.literal("<tool_call>"))));
         }
 
         // Response format parser
@@ -166,8 +167,15 @@ common_chat_params common_chat_params_init_qwen3_coder(const common_chat_templat
             auto calls      = inputs.parallel_tool_calls ? tool_call_first + p.zero_or_more(tool_call) : tool_call_first;
             auto tool_calls = p.trigger_rule("tool-call-root", p.repeat(calls, min_calls, 1));
 
+            // The templates separate content from the tool calls with one or two newlines,
+            // stop the content before them
+            std::vector<std::string> content_ends;
+            for (const auto & start : tool_call_starts) {
+                content_ends.insert(content_ends.end(), { "\n\n" + start, "\n" + start, start });
+            }
+
             return generation_prompt +
-                   (reasoning << p.content(p.until_one_of(tool_call_starts)) << tool_calls);
+                   (reasoning << p.content(p.until_one_of(content_ends)) << tool_calls);
         }
 
         // Content only parser

common/peg-parser.cpp

@@ -721,8 +721,9 @@ struct parser_executor {
                 return common_peg_parse_result(COMMON_PEG_PARSE_RESULT_SUCCESS, start_pos, pos, {}, std::move(invalid_utf8));
             }
 
-            if (match == common_trie::PARTIAL_MATCH) {
-                // Found a partial match extending to end of input, return everything before it
+            if (match == common_trie::PARTIAL_MATCH && !ctx.is_complete()) {
+                // Found a partial match extending to end of input, return everything before it.
+                // On a complete input the delimiter can no longer form, so the bytes belong to the text
                 return common_peg_parse_result(COMMON_PEG_PARSE_RESULT_SUCCESS, start_pos, pos, {}, std::move(invalid_utf8));
             }
 

common/peg-parser.h

@@ -162,9 +162,10 @@ struct common_peg_parse_result {
 };
 
 enum common_peg_parse_flags {
-    COMMON_PEG_PARSE_FLAG_NONE    = 0,
-    COMMON_PEG_PARSE_FLAG_LENIENT = 1 << 0,
-    COMMON_PEG_PARSE_FLAG_DEBUG   = 1 << 1,
+    COMMON_PEG_PARSE_FLAG_NONE     = 0,
+    COMMON_PEG_PARSE_FLAG_LENIENT  = 1 << 0,
+    COMMON_PEG_PARSE_FLAG_DEBUG    = 1 << 1,
+    COMMON_PEG_PARSE_FLAG_COMPLETE = 1 << 2,
 };
 
 inline common_peg_parse_flags operator|(common_peg_parse_flags a, common_peg_parse_flags b) {
@@ -209,6 +210,7 @@ struct common_peg_parse_context {
 
     bool is_lenient() const { return flags & COMMON_PEG_PARSE_FLAG_LENIENT; }
     bool is_debug() const { return flags & COMMON_PEG_PARSE_FLAG_DEBUG; }
+    bool is_complete() const { return flags & COMMON_PEG_PARSE_FLAG_COMPLETE; }
 };
 
 class common_peg_arena;

tests/test-chat.cpp

@@ -2473,7 +2473,7 @@ static void test_template_output_peg_parsers(bool detailed_debug) {
             .tools({
                 run_in_terminal_tool
         })
-            .expect_content("Let me inspect the current directory.\n")
+            .expect_content("Let me inspect the current directory.")
             .expect_tool_calls({
                 { "run_in_terminal", R"({"command": "pwd"})", {} },
             })
@@ -2495,7 +2495,7 @@ static void test_template_output_peg_parsers(bool detailed_debug) {
                 run_in_terminal_tool
         })
             .tool_choice(COMMON_CHAT_TOOL_CHOICE_REQUIRED)
-            .expect_content("Let me inspect the current directory.\n")
+            .expect_content("Let me inspect the current directory.")
             .expect_tool_calls({
                 { "run_in_terminal", R"({"command": "pwd"})", {} },
             })
@@ -2604,7 +2604,7 @@ static void test_template_output_peg_parsers(bool detailed_debug) {
                 .tool_choice(COMMON_CHAT_TOOL_CHOICE_REQUIRED)
                 .messages({ user_start, assistant_todos, tool_result, user_continue })
                 .expect_reasoning("I need to run a terminal command.")
-                .expect_content("Let me inspect the current directory.\n")
+                .expect_content("Let me inspect the current directory.")
                 .expect_tool_calls({
                     { "run_in_terminal", R"({"command": "pwd"})", {} },
                 })
@@ -8146,12 +8146,12 @@ static void test_chat_session() {
         for (size_t i = 0; i < thinking.size(); i += 3) {
             session.feed(common_chat_input(thinking.substr(i, 3)));
         }
-        assert_msg_equals(simple_assist_msg("", "I'm\nthinking\n"), session.msg());
+        assert_msg_equals(simple_assist_msg("", "I'm\nthinking"), session.msg());
 
         for (size_t i = thinking.size(); i < output.size(); i += 3) {
             session.feed(common_chat_input(output.substr(i, 3)));
         }
-        assert_msg_equals(simple_assist_msg("", "I'm\nthinking\n", "special_function", "{\"arg1\":1}"), session.finish());
+        assert_msg_equals(simple_assist_msg("", "I'm\nthinking", "special_function", "{\"arg1\":1}"), session.finish());
     }
 
     // a copy does not see what is fed to the original
@@ -8176,7 +8176,7 @@ static void test_chat_session() {
         common_chat_session session(tmpls.get(), nullptr, cont);
         assert_equals(std::string("<|im_start|>assistant\n<think>\nI'm thinking\n</think>\n\nHello, "),
                       session.generation_prompt());
-        assert_msg_equals(simple_assist_msg("Hello, ", "I'm thinking\n"), session.msg());
+        assert_msg_equals(simple_assist_msg("Hello, ", "I'm thinking"), session.msg());
         session.feed(common_chat_input("world!"));
         assert_equals(std::string("Hello, world!"), session.msg().content);