v0.9.0: Qwen2-VL - multi-image example?

[Katehuuh]

What example dataset do we have for multi-image? Is only supported by Qwen2-VL, not llava-1.5 or in case of (multi-)video?

From what i see in PR data/qwen2vl_demo.json and said in docs:

LLaMA-Factory/data/README.md

Line 301 in e22ac05

The number of images should be identical to the `<image>` tokens in the conversations.

So use 2nd time <image> and add in "images" can work?:

       { 
         "content": "<image>They are celebrating on the soccer field.", 
         "role": "assistant" 
       } 
     ], 
        "images": [
            "mllm_demo_data/1.jpg","mllm_demo_data/2.jpg"
        ]
   }, 

Note that in mllm_demo.json, show only 1 Img per conv:

LLaMA-Factory/data/mllm_demo.json

Lines 1 to 24 in e22ac05

	[
	{
	"messages": [
	{
	"content": "<image>Who are they?",
	"role": "user"
	},
	{
	"content": "They're Kane and Gretzka from Bayern Munich.",
	"role": "assistant"
	},
	{
	"content": "What are they doing?",
	"role": "user"
	},
	{
	"content": "They are celebrating on the soccer field.",
	"role": "assistant"
	}
	],
	"images": [
	"mllm_demo_data/1.jpg"
	]
	},

[hiyouga]

see this section: https://github.com/QwenLM/Qwen2-VL?tab=readme-ov-file#data-preparation
LLava is also supported