← sources

Maya-XS runs DeepSeek-V4.1-Flash in 120.6 GB, with all 384 experts resident in GPU memory on a 128 GB Ryzen AI Max.

source: reddit (reddit:LocalLLaMA)published 2026-10-23 11:49ingested 2026-10-24 02:40

retained text

DeepSeek-V4.1-Flash gets Maya-XS - the whole model in a 128 GB PC's memory - tensor parallel on several GPUs, and much faster drafting on one GPU.



document fetched 2026-10-24 10:43 · canonical https://reddit.com/r/LocalLLaMA/comments/1xdbtvq/mayaxs_runs_deepseekv41flash_in_1206_gb_with_all