v1

app.py

@@ -59,49 +59,53 @@ def threading_function(inputs, image_token_number, streamer, device, temperature
 @spaces.GPU
 def bot_streaming(message, history, temperature, new_max_token, top_p):
 
-    # param
-    for param in mmamba.parameters():
-        param.data = param.to(accel.device)
-    for param in meteor.parameters():
-        param.data = param.to(accel.device)
-
-    # prompt type -> input prompt
-    image_token_number = int((490/14)**2)
-    if len(message['files']) != 0:
-        # Image Load
-        image = F.interpolate(pil_to_tensor(Image.open(message['files'][0]).convert("RGB")).unsqueeze(0), size=(490, 490), mode='bicubic').squeeze(0)
-        inputs = [{'image': image, 'question': message['text']}]
-    else:
-        inputs = [{'question': message['text']}]
-
-    # [4] Meteor Generation
-    with torch.inference_mode():
-        # kwargs
-        streamer = TextIteratorStreamer(tok_meteor, skip_special_tokens=True)
-
-        # Threading generation
-        thread = Thread(target=threading_function, kwargs=dict(inputs=inputs,
-                                                               image_token_number=image_token_number,
-                                                               streamer=streamer,
-                                                               device=accel.device,
-                                                               temperature=temperature,
-                                                               new_max_token=new_max_token,
-                                                               top_p=top_p))
-        thread.start()
-
-        # generated text
-        generated_text = ""
-        for new_text in streamer:
-            generated_text += new_text
-        generated_text
-
-    # Text decoding
-    response = generated_text.split('assistant\n')[-1].split('[U')[0].strip()
+    try:
+        # param
+        for param in mmamba.parameters():
+            param.data = param.to(accel.device)
+        for param in meteor.parameters():
+            param.data = param.to(accel.device)
+
+        # prompt type -> input prompt
+        image_token_number = int((490/14)**2)
+        if len(message['files']) != 0:
+            # Image Load
+            image = F.interpolate(pil_to_tensor(Image.open(message['files'][0]).convert("RGB")).unsqueeze(0), size=(490, 490), mode='bicubic').squeeze(0)
+            inputs = [{'image': image, 'question': message['text']}]
+        else:
+            inputs = [{'question': message['text']}]
+
+        # [4] Meteor Generation
+        with torch.inference_mode():
+            # kwargs
+            streamer = TextIteratorStreamer(tok_meteor, skip_special_tokens=True)
+
+            # Threading generation
+            thread = Thread(target=threading_function, kwargs=dict(inputs=inputs,
+                                                                image_token_number=image_token_number,
+                                                                streamer=streamer,
+                                                                device=accel.device,
+                                                                temperature=temperature,
+                                                                new_max_token=new_max_token,
+                                                                top_p=top_p))
+            thread.start()
+
+            # generated text
+            generated_text = ""
+            for new_text in streamer:
+                generated_text += new_text
+            generated_text
+
+        # Text decoding
+        response = generated_text.split('assistant\n')[-1].split('[U')[0].strip()
+    
+    except:
+        response = "There are no supported something: ex) pdf, video, sound, or any other unsupported multimodal format. only single image is supported in this version."
 
     buffer = ""
     for character in response:
         buffer += character
-        time.sleep(0.02)
+        time.sleep(0.015)
         yield buffer
 
 demo = gr.ChatInterface(fn=bot_streaming,
@@ -109,6 +113,7 @@ demo = gr.ChatInterface(fn=bot_streaming,
                         additional_inputs_accordion="Generation Hyperparameters",
                         theme=gr.themes.Soft(),
                         title="☄️Meteor",
-                        description="Meteor is efficient 7B size Large Language and Vision Model built on the help of traversal of rationale",
+                        description="Meteor is efficient 7B size Large Language and Vision Model built on the help of traversal of rationale"
+                                    "Its inference speed highly depends on assinging non-scheduled GPU (Therefore, once all GPUs are busy, then inference may be taken in infinity)",
                         stop_btn="Stop Generation", multimodal=True)
 demo.launch()