This model is a 4-bit quantized version of the MNN model exported from Qwen-VL-Chat using llm-export.