Eight end-to-end Flash distillation examples across SFT, GRPO, and OPD.
The flash-example repository
contains eight small, self-contained end-to-end examples that show the full
Flash workflow. They distill Kimi K2.6, or GLM-5.2 where Kimi K2.6 was
low-yield, into 2B-9B models that match or near-match GPT-5.5 on each task’s
held-out set at a fraction of the parameter footprint. The shipped recipes span
pure SFT, single-stage OPD, SFT-to-GRPO warm starts, and SFT-to-OPD warm starts,
with exact configs and results in each example README and the repository’s
RESULTS.md.
Explore the flash-example repository
Browse the environments, training configs, held-out evaluations, and final
results.