Search results

What software is used to compare model generation speed?

Status: Open.

#82 In z-lab/dflash;

· msober opened on Apr 21, 2026

ETA for Qwen3.5-122B/397B and GLM-5.1 DFlash draft models?

Status: Open.

#81 In z-lab/dflash;

· 01lin opened on Apr 17, 2026

Question: Any timeline for open-sourcing the training recipe?

Status: Open.

#80 In z-lab/dflash;

· 01lin opened on Apr 17, 2026

Failed on AMD ROCm GPU with vLLM

Status: Open.

#72 In z-lab/dflash;

· alexhegit opened on Apr 17, 2026

MLX native is faster than the DFlash version?

Status: Open.

#71 In z-lab/dflash;

· trillionmonster opened on Apr 17, 2026

DFlash being oddly slow

Status: Open.

#70 In z-lab/dflash;

· RayBytes opened on Apr 17, 2026

Request: DFlash draft checkpoint for Qwen3.6-35B-A3B

Status: Open.

#68 In z-lab/dflash;

· afifsagent opened on Apr 16, 2026

Decode performance degrades significantly with long context

Status: Open.

#66 In z-lab/dflash;

· NASLXTO opened on Apr 16, 2026

Minimax M2.5 / M2.7

Status: Open.

#63 In z-lab/dflash;

· paxren opened on Apr 15, 2026

z-lab/Kimi-K2.5-DFlash model cannot run on vLLM.

Status: Open.

#62 In z-lab/dflash;

· lengrongfu opened on Apr 15, 2026

CUDA illegal memory access in dflash_worker_v2.py on startup and for certain requests

Status: Open.

#61 In z-lab/dflash;

· liangjason87 opened on Apr 15, 2026

Draft model acceptance collapses without thinking mode — tool-calling use cases blocked

Status: Open.

#60 In z-lab/dflash;

· FastFreddi opened on Apr 15, 2026