This model is a 4-bit quantized version of the MNN model exported from codegeex2-6b using llm-export.