implementation detail · filed under model architecture
Nemotron Hybrid Multi-Token Prediction
The Nemotron hybrid MTP method identified as a speculative-decoding configuration.
Also called nemotron_h_mtp.
- source
- 1
- model
- 1
- lab adopt it
- 1
- strongest
- optional
How sources treat it
One count per evidence span, weakest treatment to strongest.
optional 1
Documented in
Evidence
1 span quoted from the sources, strongest treatment first.
--speculative-config '{"method": "nemotron_h_mtp", "num_speculative_tokens": 5}'
optionalinference servingin Nemotron 3 UltraNVIDIA
Filed alongside
Other methods under model architecture :: prediction head.