Boost Inference Performance up to 15x on NVIDIA Blackwell Using DFlash Speculative Decoding

DFlash is an open source block diffusion model for speculative decoding that significantly accelerates LLM inference on NVIDIA Blackwell GPUs by drafting entire token blocks in parallel and verifying them efficiently with the target model. Benchmarks show DFlash delivers up to 15x throughput improvement for gpt-oss-120b and nearly doubles interactivity for Llama 3.1 8B at the same concurrency compared to EAGLE-3, with speedups of up to 5.8x for Gemma 4 31B and 5.1x for Qwen3 8-B. DFlash integrates with major inference frameworks like SGLang, vLLM, and TensorRT-LLM, allowing developers to adopt it without code refactoring.

Source: NVIDIA Developer | 2026-06-23