Intel / post

StepFun Open-Sources onPanda, Its Internal LLM Annotation and Debug Tool

Source: x.com
Date: Published Sep 22, 2026

StepFun@StepFun_ai

We’ve open-sourced onPanda 🐼 — the tool we use internally for LLM data annotation and model inspection. The workflow is simple: find an error, correct the token, and let the model continue. ✍️ Data annotation - 52% lower median annotation time vs. manual post-editing - SFT + preference data in one workflow, with high on-policy fidelity (ΔPPL <1% vs. the model’s resampling baseline) - Precise token-level supervision with paired positive/negative examples, plus agent-trajectory annotation across image, audio, and video 🔎 Model inspection and debugging - Inspect token probabilities and top-k alternatives, steer decoding token by token, and explore SVG generation, web development, and agent tasks directly in the browser. Try it (mobile-friendly): https://t.co/57vAeJetWD Paper: https://t.co/1MX1eVTREL

Sep 22, 2026 View on X

Context

Improving a often involves reviewing its outputs and producing corrected examples for further training. StepFun says it open-sourced onPanda, the tool it uses internally for LLM data annotation and model inspection. The workflow is to find an error, correct the , and let the model continue generating from there.

StepFun reports that the tool combines supervised and preference data in one workflow, with a 52% lower median annotation time than manual post-editing, and corrected traces that stay close to the model's own output (a perplexity change under 1% against the model's resampling baseline). It also lets users inspect token probabilities and top-k alternatives and steer decoding token by token in the browser, including SVG generation, web development and tasks. These are StepFun's own figures from internal use, not an independent evaluation.

Terms in this piece · Glossary

LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
fine-tuning — Taking a trained model and training it a bit more on your own examples so it gets better at one specific job.
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.

Read the source x.com