update

app.py

@@ -5,7 +5,7 @@ from threading import Thread
 import torch
 import spaces
 
-MODEL_ID = "TIGER-Lab/VL-Rethinker-7B"  
+MODEL_ID = "csfufu/Revisual-R1-final"  
 processor = AutoProcessor.from_pretrained(MODEL_ID, trust_remote_code=True)
 model = AutoModelForImageTextToText.from_pretrained(
     MODEL_ID,
@@ -14,7 +14,7 @@ model = AutoModelForImageTextToText.from_pretrained(
 ).to("cuda").eval()
 
 @spaces.GPU
-def respond(input_dict, chat_history):
+def respond(input_dict, history):
     text = input_dict["text"]
     files = input_dict["files"]
 
@@ -22,7 +22,25 @@ def respond(input_dict, chat_history):
     current_message_images = []
     messages = []
 
-    messages.append(chat_history)
+    for val in history:
+        if val[0]:
+            if isinstance(val[0], str):
+                messages.append({
+                    "role": "user", 
+                    "content": [
+                        *[{"type": "image", "image": image} for image in current_message_images],
+                        {"type": "text", "text": val[0]},
+                    ],
+                })
+                current_message_images = []
+
+            else:
+                # Load messages. These will be appended to the first user text message that comes after
+                current_message_images = [load_image(image) for image in val[0]]
+                all_images += current_message_images
+
+        if val[1]:
+            messages.append({"role": "assistant", "content": val[1]})
     
     current_message_images = [load_image(image) for image in files]
     all_images += current_message_images