Budget hardware with Qwen3.8-Flash-Next

iIIusi0n

@iIIusi0n · Sep 19

VLLM patch and guide for running Flash-Next on a CMP 170HX with streamed embeddings.

nguyenthimy2022kg-alt

@nguyenthimy2022kg-alt · Sep 8

Community vLLM build that runs the model on cheap ex-mining SM80 hardware.

polyuij42-del

@polyuij42-del · Sep 16

Cheap dual-card build that also scales to 800 tok/s at 23 concurrent users.

dg1kjd

@dg1kjd · Sep 3

Targets old V100 server cards, supporting Flash-Next and GLM-5.3-Flash.

jackjusko

@jackjusko · Sep 7

A fork tuned for Volta-generation Tesla V100 cards, giving used data-centre GPUs a path to run the 125B MoE. Author is a hobbyist maker.

moriyasujapan

@moriyasujapan · Sep 21

Reaches 1M KV tokens using FP8 PLE on a 91GB host, a cheap high-throughput setup.

MKRWW

@MKRWW · Aug 27

German-language write-up of benchmark numbers, test harness and results for running the model on three second-hand 3090s.

Shows an open-air frame of three 3060s hitting 38-40 tok/s on Strata where llama.cpp managed 13, and asks if others reuse mining hardware.