onPanda: Efficient Annotation of On-Policy Alignment Data for LLMs and Agents via Token-Level Correction
onPanda corrects the first bad token and regenerates, cutting median alignment-annotation time by 52%.
onPanda annotates LLM alignment data and agent trajectories by token-level correction: the annotator fixes the first inappropriate token, the system truncates, and generation continues from the corrected prefix. A small controlled study reports a 52% cut in median annotation time versus manual post-editing, while most tokens remain model-generated and suitable for on-policy SFT and preference data. Recorded corrections also provide position-level positive-negative pairs. The authors release the Panda-CVL dataset and a token-level correction benchmark, and the tool can connect to external harnesses.
- Annotators correct the first bad token, then generation resumes from that prefix.
- Median annotation time fell 52% versus manual post-editing in a small study.
- Outputs stay mostly on-policy and supply paired positive-negative token corrections.
- Authors release Panda-CVL and a token-level correction benchmark.
Full article191 words · extracted from huggingface.co · click to collapse
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.
Text extracted automatically; images, tables and formatting may be missing. Original: https://huggingface.co/papers/2609.24983