Incorrect list nesting in apply_chat_template causes prompts to be ignored in WISE Multimodal implementation

[boilerchun]

Describe the bug

Thanks for the work! If I understand correctly, I might find a bug in the WISE multimodal implementation.

I encountered a logic error in the code where the apply_chat_template function is called.

EasyEdit/easyeditor/models/wise/utils.py

Lines 313 to 316 in a642707

	"role": "user",
	"content": [
	[{"type": "image"}] * num_images + [{"type": "text", "text": p}]
	],

When constructing the content field for the user message, the code creates a double-nested list structure (a list inside a list: [[...]]) instead of a flat list of dictionaries.

Because the Hugging Face processor.apply_chat_template expects a flat list of content items (e.g., [{"type": "image"}, {"type": "text"}]), this extra nesting confuses the tokenizer. As a result, the actual text prompt is failed to be included in the generated prompt_ids, causing the model to receive empty or malformed inputs.

To Reproduce

... inside the apply_chat_template call ...

{
"role": "user",
# ❌ BUG: Double nesting [ [ ... ] ]
"content": [
[{"type": "image"}] * num_images + [{"type": "text", "text": p}]
],
},

... inside the apply_chat_template call ...

{
"role": "user",
# ✅ FIX: Single list [ ... ]
"content": [{"type": "image"}] * num_images + [{"type": "text", "text": p}]
},

Expected behavior

The content field should be a single, flat list containing the image placeholders and the text dictionary. When I tested, if it is not a single, flat list, the resulted prompt id will not include the actual content of the prompt.