update model weight

app.py

@@ -50,11 +50,15 @@ from torch.utils.data import DataLoader
 from pathlib import Path
 from torch.utils.data import Dataset
 import datetime
+import huggingface_hub
 
 
 device = 'cpu' 
 dtype = torch.float32
-generator = torch.load("models/generator.pt")
+MODEL_REPO = 'CVPR/generator'
+ckpt_path = huggingface_hub.hf_hub_download(
+            MODEL_REPO, f'generator.pt')
+generator = torch.load(ckpt_path)
 generator.eval()
 generator.to(device, dtype)
 params = {'batch_size': 1,
@@ -158,14 +162,14 @@ with demo:
                 The video on the right shows how the user can easily enable quantization using BigDL-Nano (with just a couple of lines of code); you may refer to our [CVPR 2022 demo paper](https://arxiv.org/abs/2204.01715) for more details.
                 ''')
         with gr.Column():
-            gr.Video(value="nano_quantize_api.mp4")
+            gr.Video(value="data/nano_quantize_api.mp4")
     gr.Markdown('''
             <h2>Demo</h2>
             
             This section uses an image stylization example to demostrate the speedup of the above code when using quantization in BigDL-Nano (about 2~3x inference time speedup). The demo is adapted from the original [FSPBT-Image-Translation code](https://github.com/rnwzd/FSPBT-Image-Translation).
             ''')
     with gr.Row().style(equal_height=False):
-        input_img = gr.Image(label="input image", value="Marvelous_Maisel.jpg", source="upload")
+        input_img = gr.Image(label="input image", value="data/Marvelous_Maisel.jpg", source="upload")
         with gr.Column():
             ori_but = gr.Button("Standard PyTorch Lightning")
             nano_but = gr.Button("BigDL-Nano")

models/generator.pt

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:975fa3b0d37bb3a019930258b52d64b8e575982069905580ac62762f2c8310d1
-size 12813658