Add zero space support.

app.py

@@ -1,3 +1,5 @@
+import spaces
+
 import os
 import shutil
 
@@ -79,6 +81,7 @@ class Chat:
         return state
 
     @torch.inference_mode()
+    @spaces.GPU
     def generate(self, tensor: list, modals: list, prompt: str, first_run: bool, state):
         # TODO: support multiple turns of conversation.
         assert len(tensor) == len(modals)

requirements.txt

@@ -4,11 +4,9 @@ torchvision==0.15.2
 torch==2.0.1
 transformers==4.37.2
 tokenizers==0.15.1
-deepspeed==0.13.1
-accelerate==0.21.0
-peft==0.4.0
-timm==0.6.13
+timm
 numpy
+spaces
 # data processing
 decord==0.6.0
 imageio==2.34.0