specific method · filed under post-training
Reinforcement learning for low-pass-rate tasks
A two-stage post-training approach that reserves reinforcement learning for tasks the model cannot yet solve at a high pass rate.
Also called RL reserved for tasks with low pass rate.
- source
- 1
- model
- 1
- lab adopt it
- 1
- strongest
- core
How sources treat it
One count per evidence span, weakest treatment to strongest.
used 1core 1
Documented in
Evidence
2 spans quoted from the sources, strongest treatment first.
The majority of what separates S 2.1 from the XS models comes from post-training, in two stages: an SFT stage that bootstraps capabilities partly with synthetic data, then RL, reserved for tasks the model can't yet solve at a high pass rate.
corepost trainingin Laguna S 2.1Poolside
then RL, reserved for tasks the model can't yet solve at a high pass rate.
usedpost trainingin Laguna S 2.1Poolside
Filed alongside
Other methods under post-training.
Scalable RL at Agent ScaleScalable reinforcement learning post-trainingSFT followed by RL and on-policy distillationExtended post-trainingJoint SFT and RLPost-training on diverse domainsPost-training optimizationThree-stage post-training with multi-teacher on-policy distillationThree-stage Thinker post-training strategy