DFlash2 authors ship Qwen3.8-Flash speculative decoding model, 20% faster than MTP at concurrency 1
rerri · reddit · 2026-10-19
Reddit user rerri shared the incoai/Qwen3.8-Flash-Next-DFlash2 model on Hugging Face, released by the original DFlash 2 authors. According to their table, it runs about 20% faster than MTP (multi-token prediction) at concurrency 1, making it notable for low-concurrency local inference setups.
More from Infra
- GPUs literally falling off the bus: MCIO riser triggers NVIDIA Xid 79 errors — TheZachMueller · 2026-10-19
- 763B DeepSeek-V4.1-Flash runs locally at 500+ tok/s on 4x RTX Pro 6000 — TheZachMueller · 2026-10-19
- selflift-Fish: 20s 1080p MiniMax H3 video on a 16GB GPU, 2.2x faster — Hot_Store_5699 · 2026-10-19
- RAM prices up 500% since 2025 — local AI blogger says buy hardware now — julianharris · 2026-10-19
- 65x speedup: 40,000-word output drops from 62.7s to 961ms — DanielLockyer · 2026-10-19
- AI Capacity Isn't Sold Out — the Biggest Deals Are — DavidLinthicum · 2026-10-19