# DFlash

12 Hacker News commenters recommended DFlash and 8 criticized it, November 2025 to November 2026. It was recommended in 6 of 13 months.

**Verdict:** Speeds up local LLM text generation, reported at 2-3x on some models, but critics say gains can vanish with higher memory use and drops in speed.

## Why people recommend it

- Boosts token generation speed (4 comments): "There are several Mac implementations of it that show > 2x faster Qwen3.5 already." ([source](https://news.ycombinator.com/item?id=47764205)) "Z Lab had been publishing dflash addons and boosting their tgen 2-3x" ([source](https://news.ycombinator.com/item?id=48866139)) "DFlash is currently a SOTA speculative decoding method" ([source](https://news.ycombinator.com/item?id=48479446))
- Enables models with no other path (2 comments): "Before this repo, there was no path to run this... SGLang's GGUF path for this model is broken." ([source](https://news.ycombinator.com/item?id=47841456)) "Before this repo, there was no path to run this" ([source](https://news.ycombinator.com/item?id=47842389))

## What people warn about

- Speed can drop off sharply (3 comments): "model performance suddenly drops off a cliff like maybe there is some memory management issue" ([source](https://news.ycombinator.com/item?id=49249443)) "performance suddenly drops off a cliff if I have DFlash enabled" ([source](https://news.ycombinator.com/item?id=49246730)) "it craters to ~9 t/s with it on, miss rate exceeding 50%" ([source](https://news.ycombinator.com/item?id=49250679))
- Higher memory use and OOM crashes (2 comments): "tgen is way up, but keep the number low, acceptance rate is terrible beyond half a dozen and it requires more memory" ([source](https://news.ycombinator.com/item?id=48849525)) "Have had issues with the dflash drafter and vllm, crashes, OOMs, broken thinking" ([source](https://news.ycombinator.com/item?id=49001956))
- Slow or no gain for some (2 comments): "Spec decode (DFlash, n-gram) didn't give a good boost" ([source](https://news.ycombinator.com/item?id=49666018)) "Dflash2 is really slow for me, like sub 10 TG" ([source](https://news.ycombinator.com/item?id=49753203))
- Broken output or failed tool calls (2 comments): "Have had issues with the dflash drafter and vllm, crashes, OOMs, broken thinking" ([source](https://news.ycombinator.com/item?id=49001956)) "DFlash2's tool call fails on python syntax." ([source](https://news.ycombinator.com/item?id=49368383))

## Alternatives hackers recommend

- [Orthrus-Qwen3](https://hackerapproved.com/p/orthrus-qwen3.md): 5 recommend

## Cite as

Hacker Approved, "DFlash", https://hackerapproved.com/p/dflash/ (counts from 1,795,640 Hacker News comments, November 2025 to November 2026).