Qwen3.6 27B DFlash — GGUF (Q4 K M + Q8 0) llama.cpp quantizations of z lab/Qwen3.6 27B DFlash , the block diffusion drafter for DFlash speculative decoding. Pair it with Qwen/Qwen3.6 27B (or a quant of it). Two quants are published: File Size Recommended? dflash draft 3.6 q8 0.gguf 1.75 GB Yes — use this. Matches F16 acceptance. dflash draft 3.6 q4 k m.gguf 1.03 GB Only if VRAM constrained; acceptance drops ~17 points. Unlike the 3.5 drafter (all full attention, Q4 robust), the 3.6 drafter introduces causal sliding window attention layers (pattern [S,S,S,S,F] , window = 2048). Those SWA layers are Q4 fragile — Q4 K M collapses acceptance from ~43 % → ~28 % on the same workload. Q8 0 is the smallest quant that preserves F16 quality and happens to run slightly faster than F16 in our benchmarks. Requirements DFlash speculative decoding is not yet in upstream llama.cpp . You need the fork: Fork: spiritbuun/buun llama cpp (branch master ) SWA support for the DFlash drafter landed in commit b9d01582b (SD 073). Older checkpoints will load the drafter but produce garbage. Built with: cmake B build DGGML CUDA=ON DGGML NATIVE=ON DGGML CUDA FA=ON DGGML CUDA FA ALL QUANTS=ON Usage llama server…
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy