4-step turbo LoRA for Qwen-Image-2.1, trained entirely on 16 Huawei Ascend 910B NPUs
Scriabinical · reddit · 2026-10-14
The LightX2V team released Qwen-Image-2.1-Lightning, a 4-step distilled LoRA for Qwen-Image-2.1 supporting text-to-image generation and image editing, trained with DMD entirely on 16 Huawei Ascend 910B NPUs. Open weights on Hugging Face; notable as a showcase of domestic Chinese NPUs for image-model distillation.
Related event: LightX2V Open-Sources 4-Step Distilled LoRA for Qwen-Image-2.1(2 posts)→
More from Infra
- Adobe ships NVIDIA RTX Spark optimizations: up to 2x faster creative apps and 5x Frame.io gains — shashib · 2026-10-16
- Meta's commitments jumped to $349B in a quarter as server memory prices tripled ahead of Oct 21 earnings — tengyanAI · 2026-10-16
- Meta releases Muse Spark Open, trained with vLLM-powered RL inference from day one — colinmcnamara · 2026-10-16
- a16z: bot 403 blocks double as Amazon cracks down on agentic commerce — a16z Newsletter · 2026-10-16
- Meta's open Muse Spark needs 493GB of memory — far beyond any consumer laptop — TejasKumar_ · 2026-10-16
- CoreWeave plunges 20% as credit markets price >60% default odds; is this dot-com all over again? — kimmonismus · 2026-10-16