@nihui

春雨为观，香月为邻 | 喜欢玩游戏时看风景 | 想成为优秀的程序员 | 看上去是女装大佬实际并不是（ | 上海老阿姨 | qwqwqwq

 Shanghai
github.com/nihui
Joined July 2008
  • Tweets6.2K
  • Following935
  • Followers44.1K
1.2KMedia
ncnn telegram chat group 技术交流电报组 ncnn community t.me/ncnnyes

ncnn community

make neural network faster :D

t.me
10
4
1
171
qwenimage-ncnn-vulkan 已支持 qwen-image-2.1-turbo 8步出图 github.com/nihui/qwenimage-n…

Release Release 20261010 · nihui/qwenimage-ncnn-vulkan

support qwen-image-2.1-turbo https://huggingface.co/nihui-szyl/qwen-image-ncnn/tree/main/qwenimage21-turbo

github.com
5
1
54
4,447
spacemit k3 pico itx
7
75
4,101
预告：4G内存 + 2G显存 跑BF16精度 qwen-image-2.1 1024x1024 出图并且速度不变慢
nihui@nihuiSep 29
用 ncnn 跑 Qwen-Image-2.1：完整 BF16、2GB 显存和全平台 Vulkan zhuanlan.zhihu.com/p/2088230… github.com/nihui/qwenimage-n…
8
29
2
269
24,210
Qwen-Image-2.1 running on Intel B580! Thanks to @nihui for qwenimage-ncnn-vulkan
1
11
985
用 ncnn 跑 Qwen-Image-2.1：完整 BF16、2GB 显存和全平台 Vulkan zhuanlan.zhihu.com/p/2088230… github.com/nihui/qwenimage-n…
🤖 Made with AI
3
31
2
290
39,558
其实没有我你分不出那些差别 结局还能多明显 别说nihui难过 别说你想改变 AI的人不用道歉😐
11
90
7,103
qwenimage-ncnn-vulkan added support for Qwen-Image-2.1 4-step LoRA and ControlNet It runs faster than pytorch-rocm or sd.cpp on my RX9060XT machine Hope my tool proves helpful to you :) github.com/nihui/qwenimage-n…
2
8
1
61
4,349
nihui retweeted
Lei Zhang
@LeiLMxSep 26
Visualizing to learn model internals in the agentic era can be surprisingly low-friction. I had Codex one-shot architecture notes earlier for several recent OSS models—DeepSeek V4/V4.1, GLM 5.3, and Kimi K3—specifically from a kernel-development perspective: tensor contracts, exact shapes/layouts, attention variants, KV-cache representation, precision choices, dataflow, and operator boundaries. Sharing in case they’re useful to others: lei.chat/models/

Model Architectures | Lei.Chat()

Interactive, source-grounded maps of modern large-model architectures.

lei.chat
5
19
1
116
10,472
好耶，libwaifu支持Qwen-Image-2.1成功🎉 github.com/ling0322/libwaifu
1
1
24
2,525
文生图性能优化 10%~20% (cooperative matrix) 图编辑性能优化 40%~85% (prefix cache) 调整低显存的 tile 策略，KV cache host swap，进一步降低显存占用，防止多图输入的显存溢出 github.com/nihui/qwenimage-n…

Release Release 20260924 · nihui/qwenimage-ncnn-vulkan

optimize image editing efficiency massively with prefix cache optimize bf16 gemm cooperative matrix kernel efficiency optimize strategy for low_vram, vae tile size and KV cache cpu staging

github.com
8
3
70
6,755
预告：controlnet 和 lora safetensor 😋
1
1,203
洗衣机里有 LVGL
9
64
5,971
Qwen-Image-2.1 full BF16 running with 2GB VRAM ncnn/Vulkan for NVIDIA/AMD/Intel/Apple No CUDA/PyTorch/Python Portable executable Text-to-image & Image editing Up to 10 reference images Transparent RGBA generation Dynamic output resolution Batch generation huggingface.co/Qwen/Qwen-Ima…

Qwen/Qwen-Image-2.1 · Qwen-Image-2.1 full BF16 running with 2GB VRAM — ncnn/Vulkan for NVIDIA, AMD,...

Hi Qwen team and everyone,

huggingface.co
19
53
4
727
50,912
预告 github.com/QwenLM/Qwen-Image…

为什么不提供最低/建议/推荐硬件规格? · Issue #4 · QwenLM/Qwen-Image-2.1

为什么不提供最低/建议/推荐硬件规格?

github.com
2
2
38
6,806
ncnn: run Cortex-M7 inference on STM32F746G-DISCO github.com/krivenkodm/embox/…

ncnn: run Cortex-M7 inference on STM32F746G-DISCO by krivenkodm · Pull Request #1 · krivenkodm/embox

STM32F746G-DISCO can now run a small NCNN network under Embox and verify its output using external SDRAM. The template starts C++/STL, allocation, and dense inference smoke tests before opening the...

github.com
1
12
2,326
I wanted to add extra GPU options to the benchmark suite, sooo I added @nihui's vkpeak to cover a range of options. Only the @RadxaComputer X4L on sbc.compare/vkpeak so far, but as with others, new boards will be tested with it, and others backfilled later! For full data, the table at the bottom of the comparison pages such as sbc.compare/radxa-x4l is where it will live.
2
1
6
762
Vulkan `vkpeak` benchmark for the new Arduino Ventuno Q reaches 400+ FP32 GFLOPS. ./vkpeak 0 device = Adreno623 driver = turnip Mesa driver / Mesa 25.2.8-0ubuntu0.24.04.2 fp32-scalar = 417.44 GFLOPS fp32-vec4 = 438.42 GFLOPS fp16-scalar = 594.99 GFLOPS fp16-vec4 = 853.58 GFLOPS int32-scalar = 109.97 GIOPS int32-vec4 = 111.09 GIOPS int16-scalar = 283.50 GIOPS int16-vec4 = 294.14 GIOPS int64-scalar = 11.31 GIOPS int64-vec4 = 6.33 GIOPS int8-dotprod = 621.29 GIOPS copy-h2h = 10.11 GBPS copy-h2d = 11.41 GBPS copy-d2h = 9.12 GBPS copy-d2d = 13.66 GBPS @arduino
1
1
4
793
YuzukiNeko
10
12
1
292
8,136
nihui retweeted
Tibo
@thsottiauxSep 7
Replying to @moonflyingoverc @twostraws
你最牛了！
472
212
120
3,970
1,386,369
nihui retweeted
柚木 鉉@GLGH_Sep 6
YuzukiNeko - A Pico Shape Cute CatBoard, With Allwinner F101 SoC Hardware OpenSource under CC0: github.com/YuzukiHD/YuzukiNe… Linux Kernel: github.com/YuzukiHD/linux-ma… Buildroot: github.com/YuzukiHD/buildroo… Zephyr: github.com/YuzukiHD/zephyr
8
42
5
437
18,216
Load more