Switch back to Docker

README.md

@@ -1,7 +1,7 @@
 ---
 title: FROMAGe
 emoji: 🧀
-sdk: gradio
+sdk: docker
 app_file: app.py
 colorFrom: blue
 colorTo: red

app.py

@@ -39,7 +39,7 @@ class FromageChatBot:
         return filename
 
 
-    def generate_for_prompt(self, input_text, state, ret_scale_factor, num_ims, num_words, temp):
+    def generate_for_prompt(self, input_text, state, ret_scale_factor, max_nm_rets, num_words, temperature):
         input_prompt = 'Q: ' + input_text + '\nA:'
         self.chat_history += input_prompt
 
@@ -50,7 +50,12 @@ class FromageChatBot:
         else:
             model_inputs = [self.chat_history]
 
-        model_outputs = self.model.generate_for_images_and_texts(model_inputs, max_num_rets=num_ims, num_words=num_words, ret_scale_factor=ret_scale_factor, temperature=temp)
+        top_p = 1.0
+        if temp != 0.0:
+            top_p = 0.95
+        model_outputs = self.model.generate_for_images_and_texts(model_inputs, 
+            num_words=num_words, ret_scale_factor=ret_scale_factor, top_p=top_p,
+            temperature=temperature, max_num_rets=max_nm_rets)
 
         im_names = []
         response = ''
@@ -77,24 +82,24 @@ class FromageChatBot:
 
 
     def launch(self):
-        with gr.Blocks(css="#fromage-space {height:600px; overflow-y:auto;}") as demo:
-            chatbot = gr.Chatbot(elem_id="fromage-space")
-            gr_state = gr.State([])
-
-            with gr.Row():
-                with gr.Column(scale=0.85):
-                    text_input = gr.Textbox(show_label=False, placeholder="Upload an image [optional]. Then enter a text prompt, and press enter!").style(container=False)
-                with gr.Column(scale=0.15, min_width=0):
-                    image_btn = gr.UploadButton("Image", file_types=["image"])     
-
-            with gr.Row():
-                with gr.Column(scale=0.20, min_width=0):
-                    clear_btn = gr.Button("Clear")
+        with gr.Blocks() as demo:
+            gr.Markdown(
+                '### Grounding Language Models to Images for Multimodal Generation'
+            )
+
+            with gr.Column(scale=0.3, min_width=0):
                 ret_scale_factor = gr.Slider(minimum=0.0, maximum=3.0, value=1.0, step=0.1, interactive=True, label="Multiplier for returning images (higher means more frequent)")
                 max_ret_images = gr.Number(minimum=0, maximum=3, value=1, precision=1, interactive=True, label="Max images to return")
                 gr_max_len = gr.Number(value=32, precision=1, label="Max # of words returned", interactive=True)
                 gr_temperature = gr.Number(value=0.0, label="Temperature", interactive=True)
 
+            with gr.Column(scale=0.7, min_width=0):
+                image_btn = gr.UploadButton("Image Input", file_types=["image"])     
+                text_input = gr.Textbox(label="Text Input", placeholder="Upload an image [optional]. Then enter a text prompt, and press enter!")
+                chatbot = gr.Chatbot()
+                gr_state = gr.State([])
+                clear_btn = gr.Button("Clear History")
+
             text_input.submit(self.generate_for_prompt, [text_input, gr_state, ret_scale_factor, max_ret_images, gr_max_len, gr_temperature], [gr_state, chatbot])
             image_btn.upload(self.upload_image, [gr_state, image_btn], [gr_state, chatbot])
             clear_btn.click(self.reset, [], [gr_state, chatbot])