Update app.py

app.py

@@ -41,38 +41,72 @@ def url_to_base64(image_url):
     else:
         return ''
 
+import json
 
 @spaces.GPU(duration=90)
 def run_inference(message, history):
-    ## may work
-    messages = []
-    images = []
-    print('\n\nmessage ',message)
-    print('\n\nhistoery ',history)
+
+    try:
+            messages= message['text']
+            print("messages   ", messages)
+            messages =  json.loads(messages)
+            final_msg=[]
+            for x in messages:
+                if x['role']=='user':
+                    tmmp=[]
+                    for y in x['content']:
+                        if y['type']=='image':
+                            tmmp+=[ImageURLChunk(image_url= url_to_base64(y['url'])]
+                        else:
+                            tmmp+=[TextChunk(text= y['text'] )]
+                    final_msg.append(UserMessage(content =tmmp   )   )
+                else:
+                    final_msg.append(AssistantMessage(content = x['content'][0]['text']   ))
     
+            completion_request = ChatCompletionRequest(messages=final_msg)
+            
+            encoded = tokenizer.encode_chat_completion(completion_request)
+            
+            images = encoded.images
+            tokens = encoded.tokens
+            
+            out_tokens, _ = generate([tokens], model, images=[images], max_tokens=512, temperature=0.45, eos_id=tokenizer.instruct_tokenizer.tokenizer.eos_id)
+            result = tokenizer.decode(out_tokens[0])
+            return result
 
-    for couple in history:
-        if type(couple[0]) is tuple:
-            images += couple[0]
-        elif couple[0][1]:
-            messages.append(UserMessage(content = [ImageURLChunk(image_url=image_to_base64(path)) for path in images]+[TextChunk(text=couple[0][1])]))
-            messages.append(AssistantMessage(content = couple[1]))
-            images = []
-    ##
         
-    messages.append(UserMessage(content = [ImageURLChunk(image_url=image_to_base64(file["path"])) for file in message["files"]]+[TextChunk(text=message["text"])]))
 
-    print('\n\nfinal messageds', messages)
-    completion_request = ChatCompletionRequest(messages=messages)
-    
-    encoded = tokenizer.encode_chat_completion(completion_request)
-    
-    images = encoded.images
-    tokens = encoded.tokens
-    
-    out_tokens, _ = generate([tokens], model, images=[images], max_tokens=512, temperature=0.45, eos_id=tokenizer.instruct_tokenizer.tokenizer.eos_id)
-    result = tokenizer.decode(out_tokens[0])
-    return result
+    ## may work
+
+    except Exception as e: 
+            messages = []
+            images = []
+            print('\n\nmessage ',message)
+            print('\n\nhistoery ',history)
+            
+
+            for couple in history:
+                if type(couple[0]) is tuple:
+                    images += couple[0]
+                elif couple[0][1]:
+                    messages.append(UserMessage(content = [ImageURLChunk(image_url=image_to_base64(path)) for path in images]+[TextChunk(text=couple[0][1])]))
+                    messages.append(AssistantMessage(content = couple[1]))
+                    images = []
+            ##
+                
+            messages.append(UserMessage(content = [ImageURLChunk(image_url=image_to_base64(file["path"])) for file in message["files"]]+[TextChunk(text=message["text"])]))
+        
+            print('\n\nfinal messageds', messages)
+            completion_request = ChatCompletionRequest(messages=messages)
+            
+            encoded = tokenizer.encode_chat_completion(completion_request)
+            
+            images = encoded.images
+            tokens = encoded.tokens
+            
+            out_tokens, _ = generate([tokens], model, images=[images], max_tokens=512, temperature=0.45, eos_id=tokenizer.instruct_tokenizer.tokenizer.eos_id)
+            result = tokenizer.decode(out_tokens[0])
+            return result
 
 demo = gr.ChatInterface(fn=run_inference, title="Pixtral 12B", multimodal=True, description="A demo chat interface with Pixtral 12B, deployed using Mistral Inference.")
 demo.queue().launch()