incoai/GLM-5.3-DFlash2

GLM-5.3-DFlash2

Blog | GitHub

This repository contains the DFlash 2 draft model for `zai-org/GLM-5.3`. It is not a standalone language model: it runs inside a speculative decoding server and drafts tokens for the target model to verify.

DFlash 2 is a block-diffusion drafter for speculative decoding. It predicts a whole block of tokens in a single pass and keeps the top candidates at every position. A lightweight selector then traces one coherent path through them. Two-tap dynamic convolutions in the backbone keep the draft from decaying toward the end of the block. Decoding is lossless: greedy output matches the target model exactly, and sampling preserves its distribution.

<div align="center"> <img src="assets/dflash2-figure.png" alt="DFlash 2: parallel block drafting with a candidate path selector" width="100%"> </div>

Quick Start

Serve with SGLang:

pip install "sglang[all] @ git+https://github.com/sgl-project/sglang.git#subdirectory=python"

sglang serve \
  --model-path zai-org/GLM-5.3 \
  --tp-size 4 \
  --trust-remote-code \
  --speculative-algorithm DFLASH \
  --speculative-draft-model-path incoai/GLM-5.3-DFlash2 \
  --speculative-draft-attention-backend fa4

DFlash 2 is also supported by vLLM v0.28.0 and later; see `incoai/GLM-5.3-NVFP4` for a vLLM serving example with the NVFP4-quantized target. See the blog post for more details.

Evaluation

• —Runtime: SGLang on four NVIDIA GB300 GPUs (TP4), with FlashAttention 4 for DFlash 2 draft attention
• —Speculation block size: 8 (7 draft tokens per verification step)
• —Sampling: GLM-5.3's officially recommended parameters (temperature 1.0, top-p 0.95), with the default Max reasoning effort
• —Maximum new tokens: 4096
• —Samples: 128 at concurrency 1; 1,024 at concurrency 8 and 32

We compare autoregressive decoding, GLM-5.3's native MTP, and DFlash 2. All speculative methods propose seven draft tokens per verification step.

Acceptance Length

Acceptance length is the per-request mean of completion tokens divided by verification steps. Higher is better.

Throughput

Throughput is total output tokens divided by end-to-end wall time. Each cell shows output tok/s (speedup vs. autoregressive).

Concurrency 1

Concurrency 8

Concurrency 32

License

This model is released under CC BY-NC-ND 4.0 for research and evaluation. For commercial licensing, contact contact@inco.ai.

Citation

If you find DFlash 2 useful, please cite:

@misc{inco2026dflash2,
  title  = {{DFlash 2: Keep Drafting Parallel}},
  author = {{Inco AI}},
  year   = {2026},
  month  = {August},
  url    = {https://inco.ai/blog/dflash2/}
}

Please also cite the original DFlash paper:

@inproceedings{chen2026dflash,
  title     = {{DFlash: Block Diffusion for Flash Speculative Decoding}},
  author    = {Chen, Jian and Liang, Yesheng and Liu, Zhijian},
  booktitle = {International Conference on Machine Learning (ICML)},
  year      = {2026}
}