What this repository contains
Retains causal exploration from real model sessions while normalizing runtime decoration. Reviewed trajectories become cumulative next-step examples for supervised fine-tuning and distillation.
MOONSHINER DATA / DATASET
Kimi K3 Traces
Kimi K3 coding, debugging, and tool-use sessions prepared for training.
THE PROJECT, IN PRACTICE
What you can do with it.
Read the schema before training
Inspect the dataset card and viewer to understand fields, split definitions, and how examples were constructed.
Preserve the intended target
Use the documented training target and provenance. Cumulative trace datasets generally supervise the final assistant step rather than repeatedly training earlier context.
FROM IDEA TO USE
How it works.
- 01
Inspect
Read the model or dataset card and its provenance.
- 02
Prepare
Check the runtime or schema and your intended use.
- 03
Explore
Open the repository and follow its current usage instructions.
greghavens/kimi-k3-coding-and-debugging-tracesUse the source page for current files, requirements, and usage examples.
TAKE IT FOR A SPIN
Start here.
Retains causal exploration from real model sessions while normalizing runtime decoration. Reviewed trajectories become cumulative next-step examples for supervised fine-tuning and distillation.
STATUS & REQUIREMENTS
Published dataset. Use its card for schema, intended use, and licensing.