adding application

app.py

@@ -4,6 +4,15 @@ from huggingface_hub import login
 # run once at startup
 if "HF_TOKEN" in os.environ:
     login(token=os.environ["HF_TOKEN"])
+
+# app.py
+import os; os.environ.setdefault('HF_HOME', '/data/hf-cache')
+os.environ.setdefault('HF_HUB_ENABLE_HF_TRANSFER', '1')
+
+from huggingface_hub import login
+login(os.getenv("HF_TOKEN", ""))
+
+from spaces import GPU
 import torch
 from exceptiongroup import catch
 from transformers import AutoTokenizer, AutoModelForCausalLM
@@ -117,6 +126,7 @@ class PatchscopesRetriever(WordRetrieverBase):
 
         return last_token_hidden_states
 
+    @GPU  # this block runs on a job GPU
     def get_hidden_states_and_retrieve_word(self, word, num_tokens_to_generate=None):
         last_token_hidden_states = self.extract_hidden_states(word)
         patchscopes_description_by_layers = self.retrieve_word(