Try PYTORCH_CUDA_ALLOC_CONF = "max_split_size_mb:512"

app.py

@@ -3,6 +3,9 @@ from peft import PeftModel
 import transformers
 import gradio as gr
 from scrape_website import process_webpages
+import os
+
+os.environ["PYTORCH_CUDA_ALLOC_CONF"] = "max_split_size_mb:512"
 
 assert (
     "LlamaTokenizer" in transformers._import_structure["models.llama"]

scrape_website.py

@@ -51,6 +51,8 @@ if __name__ == "__main__":
             urls=[
                 "https://www.example.org",
                 "https://www.example.com",
+                "https://www.imperial.ac.uk/stories/climate-action/",
+                "https://support.worldwildlife.org/site/SPageNavigator/ActionsToFightClimateChange.html",
             ]
         )
     )