fix QwQ 32B tool call parsing after thoughts (hermes2)

Author: ochafik

common/chat.cpp

@@ -1468,26 +1468,26 @@ static void common_chat_parse_hermes_2_pro(common_chat_msg_parser & builder) {
 
     static const common_regex open_regex(
         "(?:"
-        "(```(?:xml|json)?\\n\\s*)?"         // match 1 (block_start)
-        "(<tool_call>"                   // match 2 (open_tag)
-        "|<function_call>"
-        "|<tool>"
-        "|<tools>"
-        "|<response>"
-        "|<json>"
-        "|<xml>"
-        "|<JSON>"
-        ")?"
-        "(\\s*\\{\\s*\"name\"\\s*:[\\s\\S]*)"    // match 3 (named tool call + rest)
+            "(```(?:xml|json)?\\n\\s*)?" // match 1 (block_start)
+            "("                          // match 2 (open_tag)  
+                "<tool_call>"                   
+                "|<function_call>"
+                "|<tool>"
+                "|<tools>"
+                "|<response>"
+                "|<json>"
+                "|<xml>"
+                "|<JSON>"
+            ")?"
+            "(\\s*\\{\\s*\"name\"\\s*:)" // match 3 (named tool call)
         ")"
-        "|"
-        "(?:<function=([^>]+)>"            // match 4 (function name)
-        "|<function name=\"([^\"]+)\">)"  // match 5 (function name again)
-        "([\\s\\S]*)"                    // match 6 (function arguments + rest)})"
+        "|<function=([^>]+)>"            // match 4 (function name)
+        "|<function name=\"([^\"]+)\">"  // match 5 (function name again)
     );
 
+    auto start = builder.pos();
     if (auto res = builder.try_find_regex(open_regex)) {
-        if (res->groups[0].begin != 0 && res->groups[4].empty() && res->groups[5].empty()) {
+        if (res->groups[0].begin != start && res->groups[4].empty() && res->groups[5].empty()) {
             // The only syntax we allow after the very start is <function=...> or <function name=...>
             builder.add_content(builder.consume_rest());
             return;
@@ -1528,9 +1528,6 @@ static void common_chat_parse_hermes_2_pro(common_chat_msg_parser & builder) {
 
             close_tag = "</function>";
 
-            // Start parsing from after the opening tags
-            builder.move_to(res->groups[6].begin);
-
             if (auto arguments = builder.try_consume_json_with_dumped_args({{}})) {
                 if (!builder.add_tool_call(function_name, "", arguments->value) || arguments->is_partial) {
                     builder.incomplete("incomplete tool call");

tests/test-chat.cpp

@@ -894,6 +894,18 @@ static void test_template_output_parsers() {
                 "<function_call> { \"name\" : \"python\"",
                 /* is_partial= */ true,
                 {COMMON_CHAT_FORMAT_HERMES_2_PRO}));
+        assert_msg_equals(message_assist_call_thoughts,
+            common_chat_parse(
+                // QwQ-32B's template adds a trailing <think> if add_generation_prompt
+                "I'm\nthinking</think>\n"
+                "<tool_call>{\"name\": \"special_function\", \"arguments\": {\"arg1\": 1}}</tool_call>",
+                /* is_partial= */ false,
+                {
+                    /* .format = */ COMMON_CHAT_FORMAT_HERMES_2_PRO,
+                    /* .reasoning_format = */ COMMON_REASONING_FORMAT_DEEPSEEK,
+                    /* .reasoning_in_content = */ false,
+                    /* .thinking_forced_open = */ true,
+                }));
         assert_msg_equals(
             message_assist_call,
             common_chat_parse(