Model Card for Silkie

Silkie is a visual language model trained using preference distillation on GPT-4V annotated AI feedback. It is a fine-tuned version of Qwen/Qwen-VL-Chat and was trained on our MMInstruction/VLFeedback dataset with direct preference optimization (DPO). Silkie is a visual language model trained by preference distillation on GPT-4V annotated AI feedback. It is a fine-tuned version of Qwen/Qwen-VL-Chat that is trained on our MMInstruction/VLFeedback dataset with direct preference optimization (DPO). Compared with the original model, Silkile achieves 6.9% and 9.5% relative improvement on the MME benchmark regarding the perception and cognition capabilities, respectively. Besides, Silkie sets a new state-of-the-art score of 3.02 on MMHal-Bench regarding hallucination evaluation. Please refer to our project page for more details.

Model Sources

• Project page: https://vlf-silkie.github.io/
• Dataset: https://huggingface.co/datasets/MMInstruction/VLFeedback
• Paper: https://arxiv.org/abs/2312.10665
• Repository: https://github.com/vlf-silkie/VLFeedback

Uses

Silkie is intended for research purposes, particularly for alignment research in multimodal models.

How to Get Started

Below is a simple Python code snippet to get started with the model. For installation instructions please refer to our Github repository.

from transformers import AutoModelForCausalLM, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained(
    "MMInstruction/Silkie", trust_remote_code=True
)
model = AutoModelForCausalLM.from_pretrained(
    "MMInstruction/Silkie", device_map="cuda", trust_remote_code=True
).eval()
query = tokenizer.from_list_format(
    [
        {"image": "https://farm8.staticflickr.com/137/383965780_db4815011c_o.jpg"},
        {"text": "Which wooden stool has a vase with red flower on it?"},
    ]
)
response, history = model.chat(tokenizer, query=query, history=None)

Citation

@article{2023vlfeedback,
  author      = {Lei Li and Zhihui Xie and Mukai Li and Shunian Chen and Peiyi Wang and Liang Chen and  Yazheng Yang and  Benyou Wang and  Lingpeng Kong},
  title       = {Silkie: Preference Distillation for Large Visual Language Models},
  publisher   = {arXiv:2312.10665},
  year        = {2023}
}