Nex-N2.5-Max-DSpark

A DSpark draft model for speculative decoding with Nex-N2.5-Max.

This repository contains only the draft model. It is not a standalone language model: it is loaded alongside Nex-N2.5-Max and proposes several tokens per step, which the target model then verifies. The output distribution is that of Nex-N2.5-Max; the draft model only reduces latency.

Model Details


Target model	Nex-N2.5-Max
Algorithm	DSpark
Parameters	77.5B
Precision	BF16
Draft block size	5
Modality	Text

Training

This model was trained with the SpecForge library on a mix of Nex-AGI's in-house data and OpenPerfectBlend.

Deployment

Serve Nex-N2.5-Max with the draft model attached using sglang. Speculative decoding with this draft model requires our customized sglang fork, which is preinstalled in the prebuilt Docker image nexagi/sglang:v0.5.18-nex-patch:

python -m sglang.launch_server \
  --model-path <NEX_N2_5_MAX_MODEL_PATH> \
  --chat-template <NEX_N2_5_MAX_CHAT_TEMPLATE_PATH> \
  --speculative-draft-model-path <NEX_N2_5_MAX_DSPARK_MODEL_PATH> \
  --speculative-algorithm DSPARK \
  --tp-size <TP_SIZE> \
  --mem-fraction-static 0.80 \
  --host 0.0.0.0 \
  --port 30000

For the target model's own deployment options, sampling parameters, and thinking modes, see the Nex-N2.5-Max model card.