chore: removed unused pipeline

prediction.py

@@ -8,7 +8,6 @@ from xml.parsers.expat import model
 if os.environ.get("SPACES_ZERO_GPU") is not None:
     import spaces
 else:
-
     class spaces:
         @staticmethod
         def GPU(func):
@@ -67,11 +66,6 @@ class ModelPrediction:
             "{db_schema}\n"
         )
 
-    def _reset_pipeline(self, model_name):
-        if self._model_name != model_name:
-            self._model_name = model_name
-            self._pipeline = None
-
     @staticmethod
     def _extract_answer_from_pred(pred: str) -> str:
         # extract with regex everything is between <answer> and </answer>
@@ -120,24 +114,6 @@ class ModelPrediction:
             "cost": response._hidden_params["response_cost"],
         }
 
-    @spaces.GPU(duration=20)
-    def predict_with_hf(self, prompt, model_name):  # -> dict[str, Any | float]:
-        start_time = time.time()
-        outputs = pipeline(
-            [{"role": "user", "content": prompt}],
-            max_new_tokens=256,
-        )
-        end_time = time.time()
-        elapsed_time = end_time - start_time
-        # inference endpoint costs HF per Hour 3.6$/h -> 0.001 $ per second
-        cost_per_second=0.001
-        response = outputs[0]["generated_text"][-1]['content']
-        # print(response)
-        return {
-            "response": response, 
-            "cost": elapsed_time * cost_per_second
-        }
-
     def _init_model_prediction(self, model_name):
         predict_fun = self.predict_with_api
         if "gpt-3.5" in model_name: