Qwen3.6-27B speculative decoding benchmark finds DFlash fastest, with up to 4.6x speedup
thavoc77 · reddit · 2026-07-21
Benchmarks compare speculative decoding methods on Qwen3.6-27B (dense, NVFP4) on a single RTX PRO 6000 Max-Q, using the same client and three restart samples per point across vLLM and SGLang.
Main results
- DFlash is the clear winner: about 3.3x on SGLang and 2.5x on vLLM, reaching 4.6x on math reasoning.
- MTP / NEXTN delivers roughly 2.2x–2.8x, with speedups improving as draft depth increases.
- EAGLE3 peaks around 1.9x at K=3, then flattens out.
- ngram barely helps, only 1.1x–1.3x.
Engineering notes
- EAGLE3 would not load on vLLM for this model because of an hfhub head-dim validator issue; it only worked on SGLang with a patched build.
- DFlash initially crashed on the first token in SGLang because its sampler does a raw matmul on an NVFP4-quantized lmhead; dequantizing the head once fixes it.
- They also had to cap max-running-requests to avoid a mamba/GDN cache OOM.
More from Infra
- Resin lets an agent build standalone LLM inference engines; vLLM and SGLang optimize DeepSeek kernels differently — AccBalanced · 2026-10-27
- Local benchmark: Flint matches Qwen3.6 quality at 50/52 while running 3.6x faster on M4 Pro — RichardSocher · 2026-10-27
- Alchemy adds effect/Workflow support across Durable Objects, Workflows and Lambda Durable Functions — samgoodwin89 · 2026-10-27
- Alchemy's Lambda Web Functions run plain Node web servers on Lambda, no OpenNext needed — samgoodwin89 · 2026-10-27
- Foldkit Adds SSR Support With a Few Lines of Config on Cloudflare and AWS — samgoodwin89 · 2026-10-27
- Alchemy Cuts Cloudflare Bindings CPU Time by Up to 166× Via Profiling — samgoodwin89 · 2026-10-27