Update app.py

app.py

@@ -31,6 +31,7 @@ from composable_diffusion.model_creation import model_and_diffusion_defaults as
 
 has_cuda = th.cuda.is_available()
 device = th.device('cpu' if not has_cuda else 'cuda')
+print(device)
 
 # Create base model.
 timestep_respacing =  100 #@param{type: 'number'}
@@ -63,12 +64,11 @@ def show_images(batch: th.Tensor):
     reshaped = scaled.permute(2, 0, 3, 1).reshape([batch.shape[2], -1, 3])
     display(Image.fromarray(reshaped.numpy()))
 
-def compose_language_descriptions(prompt):
+def compose_language_descriptions(prompt, guidance_scale):
   #@markdown `prompt`: when composing  multiple sentences, using `|` as the delimiter.
   prompts = [x.strip() for x in prompt.split('|')]
 
   batch_size = 1
-  guidance_scale = 10 #@param{type: 'number'}
   # Tune this parameter to control the sharpness of 256x256 images.
   # A value of 1.0 is sharper, but sometimes results in grainy artifacts.
   upsample_temp = 0.980 #@param{type: 'number'}
@@ -229,13 +229,12 @@ clevr_model.to(device)
 clevr_model.load_state_dict(th.load(download_model('clevr_pos'), device))
 print('total clevr_pos parameters', sum(x.numel() for x in clevr_model.parameters()))
 
-def compose_clevr_objects(prompt):
+def compose_clevr_objects(prompt, guidance_scale):
     print(prompt)
     coordinates = [[float(x.split(',')[0].strip()), float(x.split(',')[1].strip())] 
                for x in prompt.split('|')]
     coordinates += [[-1, -1]] # add unconditional score label
     batch_size = 1
-    guidance_scale = 10
     
     def model_fn(x_t, ts, **kwargs):
         half = x_t[:1]
@@ -274,22 +273,22 @@ def compose_clevr_objects(prompt):
     return out_img
 
 
-def compose(prompt, ver):
+def compose(prompt, ver, guidance_scale):
     if ver == 'GLIDE':
-        return compose_language_descriptions(prompt)
+        return compose_language_descriptions(prompt, guidance_scale)
     else:
-        return compose_clevr_objects(prompt)
+        return compose_clevr_objects(prompt, guidance_scale)
 
 examples_1 = 'a camel | a forest'
 examples_2 = 'A cloudy blue sky  | A mountain in the horizon | Cherry Blossoms in front of the mountain'
 examples_3 = '0.1, 0.5 | 0.3, 0.5 | 0.5, 0.5 | 0.7, 0.5 | 0.9, 0.5'
-examples = [[examples_1, 'GLIDE'], [examples_2, 'GLIDE'], [examples_3, 'CLEVR Objects']]
+examples = [[examples_1, 'GLIDE', 10], [examples_2, 'GLIDE', 10], [examples_3, 'CLEVR Objects', 10]]
 
 import gradio as gr
 
 title = 'Compositional Visual Generation with Composable Diffusion Models'
 description = '<p>Demo for Composable Diffusion (~20s per example)</p><p>See more information from our <a href="https://energy-based-model.github.io/Compositional-Visual-Generation-with-Composable-Diffusion-Models/">Project Page</a>.</p><ul><li>One version is based on the released <a href="https://github.com/openai/glide-text2im">GLIDE</a> for composing natural language description.</li><li>Another is based on our pre-trained CLEVR Object Model for composing objects. <br>(<b>Note</b>: We recommend using <b><i>x</i></b> in range <b><i>[0.1, 0.9]</i></b> and <b><i>y</i></b> in range <b><i>[0.25, 0.7]</i></b>, since the training dataset labels are in given ranges.)</li></ul><p>When composing  multiple sentences, use `|` as the delimiter, see given examples below.</p>'
 
-iface = gr.Interface(compose, ['text', gr.inputs.Radio(['GLIDE','CLEVR Objects'], type="value", default='GLIDE', label='version')], 'image', title=title, description=description, examples=examples)
+iface = gr.Interface(compose, inputs=["text", gr.inputs.Radio(['GLIDE','CLEVR Objects'], type="value", default='GLIDE', label='version'), gr.Slider(1, 10)], outputs='image', title=title, description=description, examples=examples)
 
 iface.launch()