[Question] "json_invalid" when serving multimodal model #2619
Description
❓ General Questions
I saw a implement about input image in llava at https://github.com/mlc-ai/mlc-llm/pull/2208，but when i start a web serve, only text is support.
when i post under json, i can recive right result
{
"model": "HF://mlc-ai/Llama-3-8B-Instruct-q4f16_1-MLC",
"messages": [
{"role": "user", "content": "Hello! Our project is MLC LLM. What is the name of our project?"}
]
}
but when i post this json, its response "json_invalid"
{
"model": "HF://mlc-ai/Llama-3-8B-Instruct-q4f16_1-MLC",
"messages": [
{
"role": "user",
"content": [
{
"type": "image_url",
"image_url": f"{base64_encode_image('https://llava-vl.github.io/static/images/view.jpg')}",
},
{"type": "text", "text": "What does the image represent?"},
],
}
]
}