LLM inference in C/C++
C++ ggml
chat : fix gemma4 required tool grammar (#29115)
Aldehir Rojasf072b103714d
diff --git a/common/parsers/gemma4.cpp b/common/parsers/gemma4.cppindex ad48226e61..f43c5ae46d 100644--- a/common/parsers/gemma4.cpp+++ b/common/parsers/gemma4.cpp@@ -272,6 +272,10 @@ common_chat_params common_chat_params_init_gemma4(const common_chat_template &/* max = */ inputs.parallel_tool_calls ? -1 : 1));+ if (inputs.tool_choice == COMMON_CHAT_TOOL_CHOICE_REQUIRED) {+ return start + thought + tool_call;+ }+auto scan_to_toolcall = p.rule("scan-to-toolcall", p.until("<|tool_call>"));auto content = p.rule("content", p.content(p.until_one_of({"<|channel>", "<channel|>", "<|tool_call>"})));auto message = p.rule("message", thought + content);diff --git a/tests/test-chat.cpp b/tests/test-chat.cppindex 4566571e33..13733c1d9c 100644--- a/tests/test-chat.cpp+++ b/tests/test-chat.cpp@@ -3032,6 +3032,26 @@ static void test_template_output_peg_parsers(bool detailed_debug) {.expect(message_with_content_and_tool_call("Hello, world!\nWhat's up?", "get_time", R"({"city": "Paris"})")).run();+ // Required tool call+ tst.test(+ "<|tool_call>call:get_time{city:<|\"|>Paris<|\"|>}<tool_call|>")+ .tools({ get_time_tool })+ .tool_choice(COMMON_CHAT_TOOL_CHOICE_REQUIRED)+ .expect(message_with_tool_calls("get_time", R"({"city": "Paris"})"))+ .run();++ // Required tool call after reasoning+ tst.test(+ "<|channel>thought\nI'm\nthinking<channel|><|tool_call>call:get_time{city:<|\"|>Paris<|\"|>}<tool_call|>")+ .reasoning_format(COMMON_REASONING_FORMAT_AUTO)+ .tools({ get_time_tool })+ .tool_choice(COMMON_CHAT_TOOL_CHOICE_REQUIRED)+ .expect_reasoning("I'm\nthinking")+ .expect_tool_calls({+ { "get_time", R"({"city": "Paris"})", {} },+ })+ .run();+// Parallel tool callstst.test("<|tool_call>call:get_time{city:<|\"|>London<|\"|>}<tool_call|>"