Update app.py

app.py

@@ -1,5 +1,11 @@
 import gradio as gr
 from transformers import pipeline
+import torch
+
+# Set max_split_size_mb
+torch.backends.cuda.matmul.allow_tf32 = True
+torch.backends.cudnn.allow_tf32 = True
+torch.backends.cuda.matmul.max_split_size_mb = 50  # You can adjust this value
 
 raven_pipeline = pipeline(
     "text-generation",
@@ -15,6 +21,7 @@ class DialogueToSpeechConverter:
     def process_text(self, input_text: str) -> str:
         prompt = f"User Query: {input_text}<human_end>"
         result = self.raven_pipeline(prompt, max_new_tokens=2048, return_full_text=False, do_sample=False, temperature=0.001)[0]["generated_text"]
+        torch.cuda.empty_cache()
         return result
 
 # Gradio interface