Upload 3 files

app/routers/chat.py

@@ -0,0 +1,87 @@
+"""
+Endpoints الدردشة: رد كامل دفعة واحدة أو بث لحظي (SSE)
+"""
+import json
+import logging
+
+from fastapi import APIRouter, Depends
+from fastapi.responses import StreamingResponse
+
+from app.config import Settings, get_settings
+from app.dependencies import get_hf_service, verify_api_key
+from app.schemas import ChatRequest, ChatResponse, ErrorResponse
+from app.services.hf_client import HFClientError, HFInferenceService
+
+logger = logging.getLogger("hf_chat_backend.chat")
+
+router = APIRouter(
+    prefix="/chat",
+    tags=["Chat"],
+    dependencies=[Depends(verify_api_key)],
+)
+
+
+@router.post(
+    "",
+    response_model=ChatResponse,
+    responses={502: {"model": ErrorResponse}},
+    summary="إرسال محادثة والحصول على رد كامل من النموذج",
+)
+async def chat(
+    payload: ChatRequest,
+    hf_service: HFInferenceService = Depends(get_hf_service),
+    settings: Settings = Depends(get_settings),
+) -> ChatResponse | StreamingResponse:
+    if payload.stream:
+        return await chat_stream(payload, hf_service, settings)
+
+    try:
+        reply, finish_reason = await hf_service.chat_completion(
+            messages=payload.messages,
+            max_new_tokens=payload.max_new_tokens,
+            temperature=payload.temperature,
+            top_p=payload.top_p,
+        )
+    except HFClientError as exc:
+        logger.error("HF client error: %s", exc)
+        return _error_response(exc)
+
+    return ChatResponse(
+        model=settings.hf_model_id,
+        reply=reply,
+        finish_reason=finish_reason,
+    )
+
+
+async def chat_stream(
+    payload: ChatRequest,
+    hf_service: HFInferenceService,
+    settings: Settings,
+) -> StreamingResponse:
+    async def event_generator():
+        try:
+            async for delta in hf_service.chat_completion_stream(
+                messages=payload.messages,
+                max_new_tokens=payload.max_new_tokens,
+                temperature=payload.temperature,
+                top_p=payload.top_p,
+            ):
+                yield f"data: {json.dumps({'delta': delta}, ensure_ascii=False)}\n\n"
+            yield "data: [DONE]\n\n"
+        except HFClientError as exc:
+            logger.error("HF client stream error: %s", exc)
+            error_payload = json.dumps({"error": str(exc)}, ensure_ascii=False)
+            yield f"data: {error_payload}\n\n"
+
+    return StreamingResponse(event_generator(), media_type="text/event-stream")
+
+
+def _error_response(exc: HFClientError):
+    from fastapi.responses import JSONResponse
+
+    return JSONResponse(
+        status_code=502,
+        content=ErrorResponse(
+            error="hf_inference_error", detail=str(exc)
+        ).model_dump(),
+    )

app/routers/health.py

@@ -0,0 +1,19 @@
+"""
+Endpoint فحص الصحة (Health Check)
+"""
+from fastapi import APIRouter, Depends
+
+from app.config import Settings, get_settings
+from app.schemas import HealthResponse
+
+router = APIRouter(tags=["Health"])
+
+
+@router.get("/health", response_model=HealthResponse, summary="فحص صحة الخدمة")
+async def health_check(settings: Settings = Depends(get_settings)) -> HealthResponse:
+    return HealthResponse(
+        status="ok",
+        app_name=settings.app_name,
+        app_env=settings.app_env,
+        model=settings.hf_model_id,
+    )