Hy4-preview
Tencent (Hunyuan)--speculative-algorithm NEXTN --speculative-num-steps 3 --speculative-eagle-topk 1 --speculative-num-draft-tokens 4
specific method · filed under inference & serving
A serving-exposed speculative-decoding mode using a multi-step-trained MTP head or draft tokens.
The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See decoding strategy for the methods filed beside it.
Also called native NEXTN speculative decoding, NEXTN speculative decoding algorithm, NEXTN.
4 quotes from 4 sources, one card per model, each opening with its strongest.
--speculative-algorithm NEXTN --speculative-num-steps 3 --speculative-eagle-topk 1 --speculative-num-draft-tokens 4
the model ships a 1-layer MTP head trained with multiple steps, exposed by SGLang as native NEXTN speculative decoding.
--speculative-algo NEXTN --speculative-num-steps 3 --speculative-eagle-topk 1 --speculative-num-draft-tokens 4
--speculative-algo NEXTN --speculative-num-steps 3 --speculative-eagle-topk 1 --speculative-num-draft-tokens 4