Remove unused code

app.py

@@ -9,7 +9,6 @@ import pytorch_lightning as pl
 import gradio as gr
 
 from diffusers import LDMTextToImagePipeline
-# import PIL.Image
 import random
 import os
 
@@ -21,7 +20,7 @@ class ModelCheckpoint(pl.callbacks.ModelCheckpoint):
         filepath = os.path.join(self.dirpath, self.prefix + 'interrupt.ckpt')
         self._save_model(filepath)
         
-device = 'cpu' #@param ["cuda", "cpu"] {allow-input: true}
+device = 'cpu'
 reward = 'clips_grammar'
 
 cfg = f'./configs/phase2/clipRN50_{reward}.yml'
@@ -93,7 +92,7 @@ model = model.to(device)
 model.eval();
 
 import clip
-from torchvision.transforms import Compose, Resize, CenterCrop, ToTensor, Normalize
+from torchvision.transforms import Compose, Resize, CenterCrop, ToTensor
 from PIL import Image
 from timm.models.vision_transformer import resize_pos_embed
 
@@ -138,21 +137,11 @@ def generate_text_from_image(img):
         tmp_fc = tmp_fc[0]
         
         att_feat = tmp_att
-        fc_feat = tmp_fc
       
     # Inference configurations
     eval_kwargs = {}
     eval_kwargs.update(vars(opt))
   
-    verbose = eval_kwargs.get('verbose', True)
-    verbose_beam = eval_kwargs.get('verbose_beam', 0)
-    verbose_loss = eval_kwargs.get('verbose_loss', 1)
-  
-    # dataset = eval_kwargs.get('dataset', 'coco')
-    beam_size = eval_kwargs.get('beam_size', 1)
-    sample_n = eval_kwargs.get('sample_n', 1)
-    remove_bad_endings = eval_kwargs.get('remove_bad_endings', 0)
-  
     with torch.no_grad():
         fc_feats = torch.zeros((1,0)).to(device)
         att_feats = att_feat.view(1, 196, 2048).float().to(device)
@@ -162,7 +151,7 @@ def generate_text_from_image(img):
         # Only leave one feature for each image, in case duplicate sample
         tmp_eval_kwargs = eval_kwargs.copy()
         tmp_eval_kwargs.update({'sample_n': 1})
-        seq, seq_logprobs = model(
+        seq, _ = model(
             fc_feats, att_feats, att_masks, opt=tmp_eval_kwargs, mode='sample')
         seq = seq.data