BREAKING
Atomic Chat Adds DFlash Decoding
Qwen3.6-27B on RTX 6000
Baseline44
MTP65
DFlash98
0x
average gain
0tok/s
peak on JSON
0x
peak speedup
How DFlash Works
1Block-diffusion drafter
2Propose 15 tokens
3Target verifies
Strong for Code, Weak for Prose
Works wellstructured
Coding tasks
JSON output
Short context
Unreliablecreative
Long-form stories
Low acceptance
Large context windows
Speculative Decoding Goes Local
AI NEWS BLITZ
Atomic Chat has added DFlash speculative decoding, claiming much faster local inference.