This is a DFlash draft model for Qwen/Qwen3.6-35B-A3B. It is not a standalone chat or completion model. It is intended to be paired with the target model in a speculative decoding server that supports DFlash draft models and EXL3 / ExLlamaV3 quantized weights.
This is not the original BF16 Transformers checkpoint. Use it only with loaders or serving stacks that explicitly support EXL3 / ExLlamaV3 weights for this DFlash draft architecture.
If your serving stack expects the standard DFlash repository layout or BF16 weights, use the upstream model instead:
No additional fine-tuning was performed. Capabilities, limitations, license terms, and safety considerations from the upstream DFlash and target Qwen model still apply, with the usual caveat that quantization can change quality and numerical behavior.
The original upstream model card is included below for attribution and DFlash details.
This DFlash draft model is a joint retrain from Z-Lab and Modal, trained with 40k sequence length and sliding-window attention for improved long-context performance. It is mirrored across the following Hugging Face repositories:
This repository contains a DFlash draft model for Qwen/Qwen3.6-35B-A3B. It is not a standalone language model. It is intended to be paired with the target model in a speculative decoding server.
DFlash uses a lightweight block diffusion draft model to propose multiple tokens in parallel. The target model verifies those proposals, improving serving throughput while preserving the target model's output distribution.
DFlash Architecture
Quick Start
Installation
SGLang
Install a recent SGLang build with DFlash support:
uv pip install --upgrade "sglang[all]"
For best performance on Blackwell GPUs, use an SGLang build that includes DFlash, FA4/TRT-LLM attention, and FlashInfer support.
vLLM
For vLLM support, please refer to vllm-project/vllm#40898. We will update the PR to make it merge-ready soon.
Launch Server
This model should be used with an inference server that supports DFlash speculative decoding. An example SGLang deployment is:
Block size 8 is the recommended default for higher-concurrency serving. Block size 16 gives longer accept lengths and strong concurrency-1 throughput in most workloads.
Benchmark Results
We benchmarked DFlash against the autoregressive baseline and Qwen's built-in MTP draft path. DFlash reaches up to 3.61x speedup at concurrency 1 and 2.89x at concurrency 32. Across the benchmark suite, DFlash delivers higher throughput than MTP at every matched setting where both completed.
