All issues

Search results

Question about block size choice in DFlash

Status: Open.

#154 In z-lab/dflash;

· OliverZhou131 opened on Aug 13, 2026

Guidance on target hidden-layer selection

Status: Open.

#153 In z-lab/dflash;

· sleepyshep opened on Aug 10, 2026

Qwen3.5-122B-A10B drafter: 2–12% acceptance under llama.cpp with a quantized target, while Qwen3.6-35B-A3B holds 50–59% in an identical setup

Status: Open.

#152 In z-lab/dflash;

· olliehm opened on Aug 2, 2026

MLX: default block_size 16 is 24-29% below the throughput optimum on Apple Silicon

Status: Open.

#151 In z-lab/dflash;

· PhilipJohnBasile opened on Aug 1, 2026

Data point: drafts transfer across a model lineage (98% acceptance cross-model)

Status: Open.

#149 In z-lab/dflash;

· gsrunion opened on Jul 24, 2026

Request access to z-lab/MiniMax-M2.7-DFlash model

Status: Open.

#148 In z-lab/dflash;

· wangxj12 opened on Jul 13, 2026

Question about drafter for diffusion models themselves

Status: Open.

#147 In z-lab/dflash;

· hsamrat007 opened on Jul 9, 2026

[Bug] DFlash speculative decoding causes cudaErrorIllegalAddress under concurrent load when CUDA graphs are enabled

Status: Open.

#146 In z-lab/dflash;

· Love-learning-Li opened on Jul 9, 2026

Docker image is not publicly available - in contrast to README.md

Status: Open.

#144 In z-lab/dflash;

· kyr0 opened on Jun 28, 2026

KIND MODEL REQUEST: gemma4:12b-it-qat

Status: Open.

#143 In z-lab/dflash;

· ttx89-dev opened on Jun 25, 2026

NVFP4 + Dflash?

Status: Open.

#142 In z-lab/dflash;

· dude-maker opened on Jun 25, 2026

是否可以自己微调或者训练自己的dflash 呢

Status: Open.

#140 In z-lab/dflash;

· aofengdaxia opened on Jun 24, 2026