fix api bugs

src/api/routes.py

@@ -4,7 +4,7 @@ import logging
 import threading
 
 from fastapi import APIRouter, HTTPException, Request
-from fastapi.responses import StreamingResponse
+from fastapi.responses import JSONResponse, StreamingResponse
 
 from src.core.config import settings
 from src.core.engine import engine
@@ -26,14 +26,21 @@ async def chat_completions(request: Request):
     except Exception:
         raise HTTPException(status_code=400, detail="Invalid JSON")
 
-    messages = [
-        {"role": m.get("role", "user"), "content": get_clean_text(m.get("content"))}
-        for m in data.get("messages", [])
-    ]
-
-    max_tokens = data.get("max_tokens", settings.DEFAULT_MAX_TOKENS)
-    temperature = data.get("temperature", settings.DEFAULT_TEMP)
-    top_p = data.get("top_p", 0.95)
+    # Бережно собираем сообщения, сохраняя служебные поля для Tool Calling
+    messages = []
+    for m in data.get("messages", []):
+        msg = {
+            "role": m.get("role", "user"),
+            "content": get_clean_text(m.get("content")),
+        }
+        if "tool_calls" in m:
+            msg["tool_calls"] = m["tool_calls"]
+        if "tool_call_id" in m:
+            msg["tool_call_id"] = m["tool_call_id"]
+        messages.append(msg)
+
+    # Параметры от агента
+    is_stream = data.get("stream", False)
     stop = data.get("stop", [])
     if isinstance(stop, str):
         stop = [stop]
@@ -43,6 +50,30 @@ async def chat_completions(request: Request):
         if s not in stop:
             stop.append(s)
 
+    gen_kwargs = {
+        "max_tokens": data.get("max_tokens", settings.DEFAULT_MAX_TOKENS),
+        "temperature": data.get("temperature", settings.DEFAULT_TEMP),
+        "top_p": data.get("top_p", 0.95),
+        "stop": stop,
+        "stream": is_stream,
+        "tools": data.get("tools", None),
+        "tool_choice": data.get("tool_choice", None),
+    }
+
+    # Если Агент просит ответ целиком (stream=False)
+    if not is_stream:
+        loop = asyncio.get_running_loop()
+        try:
+            # Выполняем синхронный код в пуле потоков, чтобы не заблокировать FastAPI
+            response = await loop.run_in_executor(
+                None, lambda: engine.generate(messages, **gen_kwargs)
+            )
+            return JSONResponse(content=response)
+        except Exception as e:
+            logger.error(f"Sync generation error: {e}")
+            raise HTTPException(status_code=500, detail=str(e))
+
+    # Если это Чат (stream=True)
     abort_event = threading.Event()
 
     async def stream_generator():
@@ -51,18 +82,9 @@ async def chat_completions(request: Request):
 
         def worker():
             try:
-                gen_kwargs = {
-                    "max_tokens": int(max_tokens),
-                    "temperature": float(temperature),
-                    "top_p": float(top_p),
-                    "stop": stop,
-                    "abort_event": abort_event,
-                }
-
-                # Запускаем генерацию
-                for chunk in engine.generate_stream(messages, **gen_kwargs):
+                gen_kwargs["abort_event"] = abort_event
+                for chunk in engine.generate(messages, **gen_kwargs):
                     loop.call_soon_threadsafe(queue.put_nowait, chunk)
-
                 loop.call_soon_threadsafe(queue.put_nowait, None)
             except Exception as e:
                 if not abort_event.is_set():
@@ -74,7 +96,6 @@ async def chat_completions(request: Request):
         try:
             while True:
                 if await request.is_disconnected():
-                    logger.info("Client disconnected! Aborting generation...")
                     abort_event.set()
                     break
 
@@ -99,17 +120,15 @@ async def chat_completions(request: Request):
                 yield f"data: {json.dumps(chunk)}\n\n"
 
         except asyncio.CancelledError:
-            logger.info("Task cancelled. Stopping worker.")
             abort_event.set()
             raise
 
-    # Возвращаем стрим
-    headers = {
-        "X-Accel-Buffering": "no",
-        "Cache-Control": "no-cache",
-        "Connection": "keep-alive",
-        "Content-Type": "text/event-stream",
-    }
     return StreamingResponse(
-        stream_generator(), media_type="text/event-stream", headers=headers
+        stream_generator(),
+        media_type="text/event-stream",
+        headers={
+            "X-Accel-Buffering": "no",
+            "Cache-Control": "no-cache",
+            "Connection": "keep-alive",
+        },
     )

src/core/config.py

@@ -6,7 +6,7 @@ from pydantic_settings import BaseSettings
 class Settings(BaseSettings):
     # Меняем на нужную нам супер-модель
     REPO_ID: str = "empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF"
-    FILENAME: str = "Qwythos-9B-Claude-Mythos-5-1M-Q4_K_M.gguf"
+    FILENAME: str = "Qwythos-9B-Claude-Mythos-5-1M-MTP-Q5_K_M.gguf"
 
     CONTEXT_SIZE: int = 32768
     DEFAULT_MAX_TOKENS: int = 8192

src/core/engine.py

@@ -32,46 +32,56 @@ class ModelEngine:
         except Exception as e:
             print(f"CRITICAL ERROR loading model: {e}")
 
-    def generate_stream(
+    def generate(
         self,
-        messages: List[Dict[str, str]],
-        abort_event: Optional[threading.Event] = None,  # Новый аргумент
+        messages: List[Dict[str, Any]],
+        abort_event: Optional[threading.Event] = None,
         **kwargs,
-    ) -> Generator:
+    ):
         if not self.llm:
             raise RuntimeError("Model not loaded")
 
-        max_tokens = kwargs.get("max_tokens", settings.DEFAULT_MAX_TOKENS)
-        temperature = kwargs.get("temperature", settings.DEFAULT_TEMP)
-        stop = kwargs.get("stop", [])
+        stream_mode = kwargs.get("stream", True)
+
+        # Подготавливаем аргументы для llama_cpp_python
+        llama_kwargs = {
+            "messages": messages,
+            "max_tokens": int(kwargs.get("max_tokens", settings.DEFAULT_MAX_TOKENS)),
+            "temperature": float(kwargs.get("temperature", settings.DEFAULT_TEMP)),
+            "top_p": float(kwargs.get("top_p", 0.95)),
+            "stop": kwargs.get("stop", []),
+            "stream": stream_mode,
+        }
+
+        # Прокидываем инструменты (Tool Calling)
+        if kwargs.get("tools"):
+            llama_kwargs["tools"] = kwargs["tools"]
+        if kwargs.get("tool_choice"):
+            llama_kwargs["tool_choice"] = kwargs["tool_choice"]
 
         acquired = False
         while not acquired:
             if abort_event and abort_event.is_set():
                 print("Request aborted while waiting in queue.")
                 return
-
             acquired = self.lock.acquire(timeout=0.5)
 
         try:
             if abort_event and abort_event.is_set():
                 return
 
-            stream = self.llm.create_chat_completion(
-                messages=messages,
-                max_tokens=int(max_tokens),
-                temperature=float(temperature),
-                stop=stop,
-                stream=True,
-                top_p=kwargs.get("top_p", 0.95),
-            )
-
-            for chunk in stream:
-                if abort_event and abort_event.is_set():
-                    print("Request aborted during generation.")
-                    break
-
-                yield chunk
+            response = self.llm.create_chat_completion(**llama_kwargs)
+
+            # Если просят стрим - отдаем генератор
+            if stream_mode:
+                for chunk in response:
+                    if abort_event and abort_event.is_set():
+                        print("Request aborted during generation.")
+                        break
+                    yield chunk
+            # Если просят целый ответ (Агенты) - отдаем сразу словарь
+            else:
+                return response
 
         finally:
             self.lock.release()

src/ui/callbacks.py

@@ -30,6 +30,7 @@ def set_interactive(is_interactive):
     )
 
 
+# Замени bot_response на этот:
 def bot_response(history, system_prompt, temperature, max_tokens):
     messages = [{"role": "system", "content": system_prompt}]
     for msg in history[-7:]:
@@ -40,14 +41,19 @@ def bot_response(history, system_prompt, temperature, max_tokens):
     history.append({"role": "assistant", "content": ""})
 
     try:
-        stream = engine.generate_stream(
-            messages=messages, max_tokens=max_tokens, temperature=temperature
+        # Указываем stream=True для UI
+        stream = engine.generate(
+            messages=messages,
+            max_tokens=max_tokens,
+            temperature=temperature,
+            stream=True,
         )
 
         partial_text = ""
         for chunk in stream:
-            delta = chunk["choices"][0]["delta"]
-            if "content" in delta:
+            delta = chunk["choices"][0].get("delta", {})
+            # Безопасная проверка контента
+            if delta.get("content"):
                 partial_text += delta["content"]
                 history[-1]["content"] = partial_text
                 yield history