Interactive tool cuts time for fine tuning large language models by half
onPanda: Efficient Annotation of On-Policy Alignment Data for LLMs and Agents via Token-Level Correction
Computation and LanguageHuman-Computer InteractionMachine Learning
Summary
Training language models and AI agents to behave better often requires people to correct their outputs, which can be slow. The authors created onPanda, a tool that lets people fix errors token by token (words or pieces of words), then lets the model finish the response based on those fixes. This back-and-forth continues until the reply looks right, which speeds up editing compared to fixing everything after the fact. The way the tool works also helps generate better training data because it keeps most of the model’s own writing and records detailed corrections.
What this means in practice
- •For machine learning engineers: quickly produce high-quality corrected responses to train aligned language models more efficiently using token-level interactive editing.
- •For software developers building interactive ai tools: integrate token-level correction workflows to efficiently annotate user-agent interactions in realistic environments.
Authors
Lei Yang, Mengyin Liu, Jia Wang, Hangyu Guo, Liang Zhao, Zheng Ge, Kang An, Binxing Jiao, Qi Han, Daxin Jiang, Siqi Shen, Xiangyu Zhang
Abstract
We present onPanda, an interactive tool for efficiently annotating LLM alignment data and agent trajectories. onPanda adopts token-level correction as its core interaction: while reading a model response, the annotator locates the first inappropriate token and either picks a substitute from the model's candidate tokens or types the correct text via free-form editing. The system then truncates everything after that position and continues generation from the corrected prefix, repeating this locate-correct-continue loop until a satisfactory response is obtained. This mechanism lets annotators precisely steer model outputs at low cost: a small controlled study suggests that onPanda reduces median annotation time by 52% over manual post-editing. Since the vast majority of tokens in the final response are generated by the model itself, the resulting data largely preserves the model's sampling distribution and is well suited for constructing on-policy SFT and preference data. Furthermore, the token-level corrections recorded during annotation provide fine-grained supervision with precise positions and naturally paired positive--negative samples. onPanda also connects to external tools and harnesses, enabling interactive trajectory annotation in realistic environments. In addition, we release Panda-CVL, a dataset annotated with onPanda, together with a benchmark for token-level correction.