All issues

Search results

DFlashDraftModelArgs crashes on Qwen3.6-35B-A3B DFlash draft (rope_theta/block_size missing in Qwen3.6-style config)

Status: Open.

#51 In bstnxbt/dflash-mlx;

· alitrack opened on Aug 5, 2026

DFlashDraftModelArgs.from_dict() fails on config.json with nested rope_parameters/dflash_config.block_size

Status: Open.

#50 In bstnxbt/dflash-mlx;

· duffercn opened on Jul 22, 2026

Multi-model hosts: the process-global cache manager + class-level hooks limit one DFlash model per process

Status: Open.

#44 In bstnxbt/dflash-mlx;

· popfido opened on Jun 17, 2026

Feature request: prompt-lookup (n-gram) drafting as an alternative/hybrid draft source

Status: Open.

#35 In bstnxbt/dflash-mlx;

· ashalliants opened on Jun 5, 2026

Qwen3-Next target crashes in speculative rollback hook: Qwen3NextGatedDeltaNet has no attribute in_proj_qkv

Status: Open.

#33 In bstnxbt/dflash-mlx;

· JareyLi opened on May 13, 2026

M1 Max: Long-context API requests are slower with DFlash than basic decoding over Qwen3.6-35B-A3B

Status: Open.

#32 In bstnxbt/dflash-mlx;

· hdmi opened on May 11, 2026

M1 ultra 128GB issus

Status: Open.

#31 In bstnxbt/dflash-mlx;

· honeyxiaoding opened on May 10, 2026

[MLX] Low speed on Qwen3.5-27B-4bit: 15.5 tok/s with 80.9% acceptance (M4 Pro 48GB)

Status: Open.

#30 In bstnxbt/dflash-mlx;

· heykb opened on May 10, 2026

Cold-start prefill is 6-15x slower than llama.cpp Q8_0 GGUF on the same model (Apple Silicon)

Status: Open.

#21 In bstnxbt/dflash-mlx;

· hanxiao opened on May 3, 2026

CLI flags --num-draft-tokens / --decode-concurrency / --prompt-concurrency / --prefill-step-size are silently no-ops

Status: Open.

#18 In bstnxbt/dflash-mlx;

· hanxiao opened on May 3, 2026

Serve a model already downloaded.

Status: Open.

#8 In bstnxbt/dflash-mlx;

· henrique-simoes opened on Apr 19, 2026

Different Prompts achieve different results

Status: Open.

#6 In bstnxbt/dflash-mlx;

· southkorea2013 opened on Apr 17, 2026