trying to make pipeline only load once

app.py

@@ -1,8 +1,13 @@
 import gradio as gr
 from transformers import pipeline
 import transformers
+from functools import lru_cache
 
 
+@lru_cache(maxsize=4)
+def get_pipeline(task: str, model_url: str):
+    print(f"Loading pipeline: task={task}, model={model_url}")
+    return pipeline(task, model=model_url)
 
 def translate_text(task, modelUrl, chunk_to_translate):
     print(transformers.__version__)
@@ -10,7 +15,7 @@ def translate_text(task, modelUrl, chunk_to_translate):
     modelUrl = modelUrl.strip()
     task = task.strip()
 
-    pipe = pipeline(task, model=modelUrl)#these should be dynamically filled from frontend task and specific models
+    pipe = get_pipeline(task, model_url)
     result = pipe(chunk_to_translate)
     print(f"translated chunk is: {result}")
 

requirements.txt

@@ -1,4 +1,5 @@
 transformers
 gradio
 torch
-sentencepiece
+sentencepiece
+functools