Model techniques map
Techniquespost-training

specific method · filed under post-training

Three-stage post-training with multi-teacher on-policy distillation

A sequence of supervised fine-tuning, reinforcement learning for domain specialists, and multi-teacher on-policy distillation to consolidate their policies.

Also called three-stage post-training paradigm.

source
1
model
1
lab adopt it
1
strongest
core

How sources treat it

One count per evidence span, weakest treatment to strongest.

core 1

Documented in

Evidence

1 span quoted from the sources, strongest treatment first.

Our post-training pipeline follows a three-stage paradigm: initializing baseline agent capabilities via supervised fine-tuning (SFT), developing specialized domain experts at varying reasoning effort via Reinforcement Learning (RL), and consolidating these domain-specific policies into a single model using Multi-Teacher On-Policy Distillation (MOPD).

coreunclearin Kimi K3Moonshot AI

Filed alongside

Other methods under post-training.