Feats: Update inference

app.py

@@ -1,6 +1,7 @@
 import argparse
 import gradio as gr
 import os
+import spaces
 from PIL import Image
 
 from kimi_vl.serve.frontend import reload_javascript
@@ -89,6 +90,7 @@ def get_prompt(conversation) -> str:
 
 
 @wrap_gen_fn
+@spaces.GPU()
 def predict(
     text,
     images,

kimi_vl/serve/inference.py

@@ -153,7 +153,7 @@ def preprocess(
     return inputs
 
 
-@spaces.GPU
+
 def kimi_vl_generate(
     model: torch.nn.Module,
     processor: AutoProcessor,
@@ -181,7 +181,6 @@ def kimi_vl_generate(
     )
 
 
-@spaces.GPU
 def generate(
     model,
     processor,