Update app.py

app.py

@@ -8,7 +8,7 @@ app = FastAPI()
 
 hf_hub_download("TheBloke/deepseek-coder-1.3b-instruct-GGUF", "deepseek-coder-1.3b-instruct.Q5_K_M.gguf", local_dir="./")
 
-model_l = Llama(model_path="./deepseek-coder-1.3b-instruct.Q5_K_M.gguf", n_ctx=16000, n_gpu_layers=0, n_threads=2, use_mlock = True)
+model_l = Llama(model_path="./deepseek-coder-1.3b-instruct.Q5_K_M.gguf", n_ctx=16000, n_gpu_layers=0, n_threads=2, use_mlock=True)
 
 @app.get("/check")
 async def index():
@@ -17,12 +17,12 @@ async def index():
 @app.post("/api")
 async def completion(request: Request):
     data = await request.json()
-    prompt = data["prompt"]
+    prompt = f"Complete the following code, do not use comments too much and longer than half-line(return only the completed code!):\n\n{data["prompt"]}"
     res = model_l(
         prompt,
-        temperature=0.5,
+        temperature=0.7,
         echo=False,
-        max_tokens=32,
+        max_tokens=64,
     )
     return {"responses": res["choices"]}