ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-02-07 23:10:10 +00:00
Code Issues Packages Projects Releases Wiki Activity

Enable LLM function calls (#643)

Browse Source
This commit is contained in:
Anton Sokolchenko
2025-07-24 20:24:12 +02:00
committed by GitHub
parent dffa0a95b3
commit cfc8f5a61b
2 changed files with 43 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
examples/server/server.cpp
View File

@@ -1683,6 +1683,7 @@ struct server_context {
res.stop = true;
res.data = json {
{"content", !slot.params.stream ? slot.generated_text : ""},
{"generated_text", slot.generated_text}, // Always include full text for finish_reason logic
{"id_slot", slot.id},
{"stop", true},
{"model", params.model_alias},
@@ -2822,11 +2823,22 @@ static std::vector<json> format_partial_response_oaicompat(server_task_result ta
std::string content = json_value(result, "content", std::string(""));
std::string finish_reason;
if (stopped_word || stopped_eos) {
finish_reason = "stop";
}
if (stopped_limit) {
finish_reason = "length";
} else if (stopped_word || stopped_eos) {
// Following original llama.cpp pattern: finish_reason = oaicompat_msg.tool_calls.empty() ? "stop" : "tool_calls"
// Use generated_text (complete content) for finish_reason logic, not content (empty in streaming)
std::string generated_text = json_value(result, "generated_text", std::string(""));
ik_chat_msg final_msg = parse_chat_message_incremental(generated_text, false, modelname);
// Debug logging
LOG_INFO("DEBUG: Streaming finish_reason check", {
{"generated_text", generated_text},
{"model_name", modelname},
{"tool_calls_count", final_msg.tool_calls.size()}
});
finish_reason = final_msg.tool_calls.empty() ? "stop" : "tool_calls";
}
std::time_t t = std::time(0);