The released ckpt cannot reproduce the experimental results

[zengkaiya]

Paper Results

Reproduce the results

[xinwei666]

Thank you for your attention. Last week, we fixed a bug in the training code, retrained a model, and uploaded it to the previous Hugging Face repository. This week, we organized and uploaded the inference and model-based evaluation scripts. We tested the current version of the model, and the overall Strict/Relaxed Accuracy scores are 25.07/52.07.

[zengkaiya]

Thank you for your efforts. I will retest and give you feedback.🌹