← sources

club-3090 / qengine / RedHatAI/Qwen3.8-27B-INT4

source: reddit (reddit:LocalLLaMA)published 2026-10-29 14:53ingested 2026-10-30 02:40

retained text

This thing [https://github.com/noonghunna/club-3090/blob/master/models/qwen3.8-27b/qengine/compose/dual/redhat-int4/mtp.yml](https://github.com/noonghunna/club-3090/blob/master/models/qwen3.8-27b/qengine/compose/dual/redhat-int4/mtp.yml)

[https://github.com/bpx03/QEngine](https://github.com/bpx03/QEngine)

[https://huggingface.co/RedHatAI/Qwen3.8-27B-INT4#accuracy](https://huggingface.co/RedHatAI/Qwen3.8-27B-INT4#accuracy)

i'm on 2x3090, and dual VLLM qwen3.8-27b fp8 was my daily.  some of my buddies were yakkin about local AI so i figured i'd try another preset.  HOLY HELL is it way faster than fp8! (i mean of course it will be) but its SIGNIFICANTLY faster than i remember the last int4 i tried when qwen3.8-27b first came out.

have any of yall kicked the tires on this?

https://preview.redd.it/nnns4wsq6fyh1.png?width=913&format=png&auto=webp&s=db487304d8d7b3e1fd1dc0c62173793e6050f033

document fetched 2026-10-30 00:44 · canonical https://reddit.com/r/LocalLLaMA/comments/1xim690/club3090_qengine_redhataiqwen3827bint4