This model is a 4-bit quantized version of the MNN model exported from Qwen1.5-4B-Chat using llm-export.