Server: add support for "tool_calls" (MeetKai/functionary model)

.gitignore

@@ -92,3 +92,4 @@ examples/jeopardy/results.txt
 poetry.lock
 poetry.toml
 nppBackup
+functionary-test

Makefile

@@ -719,7 +719,7 @@ save-load-state: examples/save-load-state/save-load-state.cpp ggml.o llama.o $(C
 	$(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<)
 	$(CXX) $(CXXFLAGS) $(filter-out %.h $<,$^) $(call GET_OBJ_FILE, $<) -o $@ $(LDFLAGS)
 
-server: examples/server/server.cpp examples/server/oai.hpp examples/server/utils.hpp examples/server/httplib.h examples/server/json.hpp examples/server/index.html.hpp examples/server/index.js.hpp examples/server/completion.js.hpp examples/llava/clip.cpp examples/llava/clip.h examples/llava/llava.h examples/llava/llava.cpp common/stb_image.h ggml.o llama.o $(COMMON_DEPS) grammar-parser.o $(OBJS)
+server: examples/server/server.cpp examples/server/oai.hpp examples/server/utils.hpp examples/server/httplib.h examples/server/json.hpp examples/server/functionary.hpp examples/server/index.html.hpp examples/server/index.js.hpp examples/server/completion.js.hpp examples/llava/clip.cpp examples/llava/clip.h examples/llava/llava.h examples/llava/llava.cpp common/stb_image.h ggml.o llama.o $(COMMON_DEPS) grammar-parser.o $(OBJS)
 	$(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<)
 	$(CXX) $(CXXFLAGS) -c examples/llava/clip.cpp -o $(call GET_OBJ_FILE, examples/llava/clip.cpp) -Wno-cast-qual
 	$(CXX) $(CXXFLAGS) -Iexamples/server $(filter-out %.h %.hpp $< examples/llava/clip.cpp,$^) $(call GET_OBJ_FILE, $<) $(call GET_OBJ_FILE, examples/llava/clip.cpp) -o $@ $(LDFLAGS) $(LWINSOCK2)

examples/server/CMakeLists.txt

@@ -1,7 +1,7 @@
 set(TARGET server)
 option(LLAMA_SERVER_VERBOSE "Build verbose logging option for Server" ON)
 include_directories(${CMAKE_CURRENT_SOURCE_DIR})
-add_executable(${TARGET} server.cpp oai.hpp utils.hpp json.hpp httplib.h)
+add_executable(${TARGET} server.cpp oai.hpp utils.hpp json.hpp functionary.hpp httplib.h)
 install(TARGETS ${TARGET} RUNTIME)
 target_compile_definitions(${TARGET} PRIVATE
     SERVER_VERBOSE=$<BOOL:${LLAMA_SERVER_VERBOSE}>

examples/server/functionary-test.cpp

@@ -0,0 +1,110 @@
+#include <iostream>
+#include <string>
+#include <vector>
+#include <sstream>
+
+#include "json.hpp"
+#include "functionary.hpp"
+
+using json = nlohmann::json;
+
+/**
+ * A simple test program that allow testing functionary.hpp without using server.
+ * TODO: how to add this test to CI?
+ * 
+ * Compile command: clear && g++ functionary-test.cpp -o functionary-test && ./functionary-test
+ */
+
+std::string test_oai_input_json = R"(
+{
+    "tools": [
+        {
+            "type": "function",
+            "function": {
+                "name": "get_n_day_weather_forecast",
+                "description": "Get an N-day weather forecast",
+                "parameters": {
+                    "type": "object",
+                    "properties": {
+                        "location": {
+                            "type": "string",
+                            "description": "The city and state, e.g. San Francisco, CA"
+                        },
+                        "format": {
+                            "type": "string",
+                            "enum": ["celsius", "fahrenheit"],
+                            "description": "The temperature unit to use. Infer this from the users location."
+                        },
+                        "num_days": {
+                            "type": "integer",
+                            "description": "The number of days to forecast"
+                        }
+                    },
+                    "required": ["location", "format", "num_days"]
+                }
+            }
+        }
+    ],
+    "messages": [
+        {
+            "role": "system",
+            "content": "You are a helpful assistant"
+        },
+        {
+            "role": "user",
+            "content": "What is the weather like in Boston?"
+        },
+        {
+            "role": "assistant",
+            "content": null,
+            "tool_calls": [
+                {
+                    "type": "function",
+                    "id":"get_car_price",
+                    "function": {
+                        "arguments": "{\"car_name\": \"Song\"}",
+                        "name": "get_car_price"
+                    }
+                },
+                {
+                    "type": "function",
+                    "id":"get_car_price",
+                    "function": {
+                        "arguments": "{\"car_name\": \"Tang\"}",
+                        "name": "get_car_price"
+                    }
+                }
+            ]
+        },
+        {
+            "role": "tool",
+            "tool_call_id": "get_car_price",
+            "name": "get_car_price",
+            "content": "{\"price\": {\"price\": \"$25000\"}}"
+        },
+        {
+            "role": "tool",
+            "tool_call_id": "get_car_price",
+            "name": "get_car_price",
+            "content": "{\"price\": {\"price\": \"$20000\"}}"
+        }
+    ]
+}
+)";
+
+
+std::string test_response = R"(get_car_price
+<|content|>{"car_name": "Song"}
+<|from|>assistant
+<|recipient|>get_car_price
+<|content|>{"car_name": "Tang"}<|stop|>)";
+
+int main() {
+    auto test_oai_input = json::parse(test_oai_input_json);
+    auto prompt = llama_functionary::convert_oai_to_prompt(test_oai_input, true);
+    std::cout << "\n" << prompt << "\n";
+
+    std::cout << "\n" << llama_functionary::convert_response_to_oai_choices(test_response) << "\n";
+
+    return 0;
+}