Deploy a local large model Implement automatic splitting and content recognition of PDF documents (model fine-tuning is required) Output the recognition results in JSON format Specify using python Recognition accuracy is the core requirement, source code delivery after passing testing