Fix lazy trigger handling

Author: Olivier Chafik

examples/server/server.cpp

@@ -402,8 +402,7 @@ struct server_task {
         {
             const auto grammar_trigger_words = data.find("grammar_trigger_words");
             if (grammar_trigger_words != data.end()) {
-                auto words = to_string_vec(*grammar_trigger_words);
-                for (const auto & word : params.sampling.grammar_trigger_words) {
+                for (const auto & word : to_string_vec(*grammar_trigger_words)) {
                     auto ids = common_tokenize(vocab, word, /* add_special= */ false, /* parse_special= */ true);
                     if (ids.size() == 1) {
                         params.sampling.grammar_trigger_tokens.push_back(ids[0]);

examples/server/tests/unit/test_chat_completion.py

@@ -273,12 +273,12 @@ def test_completion_with_required_tool(template_name: str, n_predict: int, tool:
 
 
 @pytest.mark.parametrize("template_name,n_predict,tools,tool_choice", [
-    ("meetkai-functionary-medium-v3.1",               32,  [],            None),
-    ("meetkai-functionary-medium-v3.1",               32,  [TEST_TOOL],   None),
-    ("meetkai-functionary-medium-v3.1",               32,  [PYTHON_TOOL], 'none'),
-    ("meetkai-functionary-medium-v3.2",               32,  [],            None),
-    ("meetkai-functionary-medium-v3.2",               32,  [TEST_TOOL],   None),
-    ("meetkai-functionary-medium-v3.2",               32,  [PYTHON_TOOL], 'none'),
+    ("meetkai-functionary-medium-v3.1",               128, [],            None),
+    ("meetkai-functionary-medium-v3.1",               128, [TEST_TOOL],   None),
+    ("meetkai-functionary-medium-v3.1",               128, [PYTHON_TOOL], 'none'),
+    ("meetkai-functionary-medium-v3.2",               128, [],            None),
+    ("meetkai-functionary-medium-v3.2",               128, [TEST_TOOL],   None),
+    ("meetkai-functionary-medium-v3.2",               128, [PYTHON_TOOL], 'none'),
     ("meta-llama-Meta-Llama-3.1-8B-Instruct",         128, [],            None),
     ("meta-llama-Meta-Llama-3.1-8B-Instruct",         128, [TEST_TOOL],   None),
     ("meta-llama-Meta-Llama-3.1-8B-Instruct",         128, [PYTHON_TOOL], 'none'),