Improve the Qwen25VLChatHandler's ability to handle multiple image inputs to avoid the illusion of multiple image inputs

Author: JamePeng

llama_cpp/llama_chat_format.py

@@ -3634,7 +3634,7 @@ class Qwen25VLChatHandler(Llava15ChatHandler):
     DEFAULT_SYSTEM_MESSAGE = "You are a helpful assistant."
 
     CHAT_FORMAT = (
-        #"{% set image_count = namespace(value=0) %}"
+        "{% set image_count = namespace(value=0) %}"
         #"{% set video_count = namespace(value=0) %}"
         "{% for message in messages %}"
         "{% if loop.first and message['role'] != 'system' %}"
@@ -3648,11 +3648,12 @@ class Qwen25VLChatHandler(Llava15ChatHandler):
         "{% for content in message['content'] %}"
         "{% if content['type'] == 'image_url' %}"
         "{% if content.image_url is string %}"
-        "{{ content.image_url }}"
+        "{% set image_count.value = image_count.value + 1 %}"
+        "Picture {{ image_count.value }}: <|vision_start|> {{ content.image_url }} <|vision_end|>"
         "{% else %}"
-        "{{ content.image_url.url }}"
+        "{% set image_count.value = image_count.value + 1 %}"
+        "Picture {{ image_count.value }}: <|vision_start|> {{ content.image_url.url }} <|vision_end|>"
         "{% endif %}"
-        #"{% set image_count.value = image_count.value + 1 %}"
         "{% elif content['type'] == 'text' %}"
         "{{ content['text'] }}"
         "{% endif %}"