Model techniques map
Techniquesinference & servingdecoding strategy

implementation detail · filed under inference & serving

Distribution-matched draft-model fine-tuning

Fine-tunes a speculative draft model on early RL rollout logs resampled to match the RL training distribution.

source
1
models
2
lab adopt it
1
strongest
used

How sources treat it

One count per evidence span, weakest treatment to strongest.

used 1

Documented in

Evidence

1 span quoted from the sources, strongest treatment first.

DFlash is then finetuned on early RL rollout logs resampled to match the RL training distribution.

usedinference servingin MiMo-V2.6Xiaomi

Filed alongside

Other methods under inference & serving :: decoding strategy.