Inco Open-Sources DFlash 2 Drafter: Lossless 3.5× Speedup for MiniMax M3.1 Flash
Lianhuiq · x · 2026-10-23
Inco AI released an Apache 2.0 DFlash 2 draft model for MiniMax M3.1 Flash, delivering up to 3.5× faster generation with no quality loss — greedy output matches the target model exactly and sampling preserves its distribution.
Technical highlights: DFlash 2 is a block-diffusion drafter that predicts a whole block of tokens per pass, keeps top candidates at every position, and uses a lightweight selector to trace one coherent path; two-tap dynamic convolutions prevent decay toward the block's end. The model card includes SGLang config (8-token draft blocks, fa4 backend), benchmarked on 2×GB300 (TP2/EP2) with MXFP8 weights and an NVFP4 sparse KV cache.
More from Infra
- NVIDIA's $1.2M HGX Rubin NVL8 hits GTC Berlin, promising 10x tokens per megawatt over B200 — thursdai_pod · 2026-10-23
- Amazon scraps plans for four data centers on $427M former GWU Ashburn campus — GaryMarcus · 2026-10-23
- Developer platforms raided for AI inference keys: Railway, Fireworks, LiteLLM all hit this month — ccerrato147 · 2026-10-23
- Koi Paper: Self-Calibrating Planner Boosts Heterogeneous LLM Fleet Goodput 1.18x, Cuts $/token 20% — burny_tech · 2026-10-23
- Cloudflare CEO: agent traffic will be 1000x human traffic in 5 years, internet needs micropayments — 4KTV · 2026-10-23
- vLLM details M3.1 Flash serving: NVFP4 cuts weights 44.6%, boosts capacity 2.7x — vllm_project · 2026-10-23