Paragraph Handling Regression
Paragraph Handling Regression: a task in harbor-swesmith-rl-artifacts: Harbor SWE-Smith 强化学习数据产物 (Harbor dataset). It seems there's a regression in how paragraphs are being handled in the latest update. Previously, when adding a paragraph, the text would be appended to the last paragraph token if…
The task
It seems there's a regression in how paragraphs are being handled in the latest update. Previously, when adding a paragraph, the text would be appended to the last paragraph token if it existed. However, now each paragraph is being treated as a separate token, which is causing unexpected behavior in the output.