Spaces:
Running on L40S
Differences between Gradio and code output
Hello everyone,
I wanted to test MinerU on my server, so I installed everything inside a Docker starting from the latest slang image, and I got everything working. I copied the parameters that your demo dashboard on Gradio is sending, and I'm using the vlm_slang. However, I get very different results loading the same document in your dashboard or in my code. Am I missing something? is the open-source code a different version from the demo one?
I call in this way, with method "auto", language "ch", and backend "vlm-sglang-engine"
Thank you!
Please commit more information about the different.
Hi,
I am having the same issue with the docker implementation and the demo. The demo produces the correct results but the docker implementation is off e.g it is capturing the header but the row data is missing.
I know it is not the OCR model as I updated to MinerU2.5-Pro-2604-1.2B in docker.
These are the parameters that I am using:
"return_middle_json": "false",
"return_model_output": "false",
"return_md": "true",
"return_images": "false",
"end_page_id": "99999",
"parse_method": "auto",
"start_page_id": "0",
"lang_list": "en",
"server_url": "",
"return_content_list": "false",
"backend": "vlm-auto-engine",
"table_enable": "true",
"response_format_zip": "false",
"return_original_file": "false",
"formula_enable": "true",
Thanks. I realize I had to update to the latest model opendatalab/MinerU2.5-Pro-2604-1.2B on mineru-api.
I was updating model on the wrong instance on mineru-openai-server
I am now getting the same results in my docker deploy versus the online demo