Our #TextToImage inference engine #OneDiff has just updated to a new version, which allows you to change the image size for most models without recompilation. Additionally, OneDiff ComfyUI nodes are integrated into the ComfyUI-Manager. #AIGC#GenerativeAIreddit.com/r/StableDiffusion…
Hello world! We are the #SiliconFlow team dedicated to building #AI infra. We currently focus on high-performance and cost-saving large-scale model inference and serving solutions. More updates from us coming on X. Stay tuned:-) #LLM#GenerativeAI#TextToImage
另外我想推荐一个中文公众号OneFlow,链接是他们经授权后翻译的 Eye on AI 跟 ILYA SUTSKEVER 的对谈。虽然 OneFlow是中国一家做ML框架和实训社区的纯技术公司,我也是意外发现他们的媒体团队应该是中文自媒体里采编译水平最高的,无论内容源还是翻译质量,技术方面也很过硬
mp.weixin.qq.com/s/rZBEDlxFV…
We are thrilled to announce the release of #OneFlow v0.9.0. This update contains 640 commits. For the full changelog, please check out: github.com/Oneflow-Inc/onefl…. Come install OneFlow v0.9.0 for a new user experience. Your feedback will be much appreciated!
oneflow2020.medium.com/onefl…
What's the journey of an operator in a deep learning framework? This article will take output = flow.relu(input) as an example to illustrate how an op is executed from Python to C in OneFlow. #AI#DeepLearninglink.medium.com/fQ7L6J6U6qb
This article will unveil the mathematical property of collective communication operations by analyzing the case of all-reduce, which is common in data parallelism link.medium.com/rHedRE02Tqb
Activation function PReLU is frequently used in InsightFace. In this article, we are talking about OneFlow's way of optimizing PPeLU in CUDA.
link.medium.com/nUK7SLMOfqb
In this series, we will discuss why need to build a new point-to-point communication library, what an ideal point-to-point communication library should be, and how to work together to construct an open-source project for it.
link.medium.com/b0g1cGbZRlb
This article lists papers on GShard, presents background information and inspiration from the papers, and finally evaluates what else can be done to improve GShard from similar work that has been done in OneFlow
link.medium.com/qsNlLHarGlb