Xformers comfyui

ops. py is also heavily used. Update the server. The most powerful and modular stable diffusion GUI, api and backend with a graph/nodes interface Aug 5, 2023 · How to Update ComfyUI xformers. x, SD2. ComfyUIはStable Diffusionモデルから画像を生成する、Webブラウザベースのツールです。. This is what I deal with in webui when xformers is on: (Here is a me as an examplee) It destroys images. [1] ComfyUI looks Feb 9, 2024 · edited. First go to the Xformers python library official release on Pypi website. 1. 17 for systems with torch 1. 制作了中文版ComfyUI插件与节点汇总表，项目详见：【腾讯文档】ComfyUI 插件（模组）+ 节点（模块）汇总 【Zho】 20230916 近期谷歌Colab禁止了免费层运行SD，所以专门做了Kaggle平台的免费云部署，每周30小时免费冲浪时间，项目详见： Kaggle ComfyUI云部署1. 16 of xFormers, released on January 2023, installation can be easily performed using pre-built pip wheels: Sep 18, 2023 · It's anxformers bug accidentally triggered by the way the original AnimateDiff CrossAttention is passed in. 2+ performs well on Windows and is more consistent (albeit subtly), so it’s understandable why the ComfyUI official package chose not to include xFormers. post1. Launch ComfyUI by running python main. First, locate the folder where ComfyUI’s xformers package is installed. then! ## insatll xformers match your torch,for torch==2. bat to start ComfyUI! Alternatively you can just activate the Conda env: python_miniconda_env\ComfyUI, and go to your ComfyUI root directory then run command python . 0. 🎨ComfyUI standalone pack with 30+ custom nodes. 389605. I've got 16GB of VRAM (on a 2080), and normal mode uses half of it but only while working on an image. Thanks for ControlAltAI in youtube and Kijai in github. 357721 ** Platform: Windows ** Python version: 3. We read every piece of feedback, and take your input very seriously. 24 Set vram state to: NORMAL_VRAM File "D:\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-Moore-AnimateAnyone_init. 13. xformers 早就在用了，看你有没设置启动而已。. py", line 65, in _run_priority_list raise NotImplementedError(msg) The text was updated successfully, but these errors were encountered: Oct 18, 2023 · Xformers builds really easily on Windows for pytorch 2. Navigating the ComfyUI User Interface. Note: Remember to add your models, VAE, LoRAs etc. Faster examples with accelerated inference. Feb 17, 2024 · Saved searches Use saved searches to filter your results more quickly Added an example of efficient LLaMa decoding using xformers operators. Additionally you could instead use --opt-sdp-no-mem-attention if you want deterministic rendering (same seed produces exactly the same results every time) 4. Jan 24, 2024 · Deploy a fresh Ubuntu 22. format: supports image/gif, image/webp (better compression), video/webm, video/h264-mp4, video/h265-mp4. Feb 22, 2024 · Lets see how we can upgrade it. 23. Download Link with unstable nightly pytorch. Download files. 0, which has no compatible version of xformers, so the log complained that I had no xformers installed. If you’re using ComfyUI and need to update the xformers package to ensure you have the latest features and improvements, follow these steps: Step 1: Locate the xformers folder. Mar 26, 2023 · Create your ComfyUI venv (like in the cloned repo) Ensure that venv is active; Do what's needed to install pytorch & xformers according to your platform (e. 0 Token merging →. on Aug 12, 2023. warn ("DWPose: Onnxruntime not found or doesn't come with acceleration providers, switch to OpenCV with CPU device. Contribute to AIFSH/ComfyUI-Hallo development by creating an account on GitHub. Jul 1, 2023 · Run the following: python setup. py Collaborate on models, datasets and Spaces. or if you use portable (run this in ComfyUI_windows_portable -folder): python_embeded\python. 6 (tags/v3. • 1 yr. In stable-diffusion-webui directory, install the . dustysys/ ddetailer - DDetailer for Stable-diffusion-webUI extension. And above all, BE NICE. 2. 17+c36468d. When Im trying to generate image and process going thru KSammpler i receiving an Mar 2, 2024 · On a fresh ComfyUI install, whenever I run the script, it seems to load the FrozenCLIPembedder, and then stop the script. Mar 27, 2023 · I do have xformers active, I think, in comfyui (default settings for portable). 25. Please keep posted images SFW. Added Flash-Decoding for faster attention during Large Language Model (LLM) decoding - up to 50x faster for long sequences (token decoding up to 8x faster end-to-end) Added an efficient rope implementation in triton, to be used in LLM decoding. 今回は、ControlNetの「Tile」という機能を組み合わせて、2枚の Thank you, I deleted all xformers folders out of ~\ComfyUI_windows_portable\python_embeded\Lib\site-packages (I had two xformers folders in there, xformers and xformers-0. We would like to show you a description here but the site won’t allow us. to the corresponding Comfy folders, as discussed in ComfyUI manual installation. xformers is automatically detected and enabled if found, but it's not necessary, in some cases it can be a bit faster though: pip install -U xformers --no-dependencies (for portable python_embeded\python. Aug 8, 2023 · 2. A modal appears with the below text: 1 day ago · Copy the files inside folder __New_ComfyUI_Bats to your ComfyUI root directory, and double click run_nvidia_gpu_miniconda. What could be causing this? Total VRAM 24564 MB, total RAM 32693 MB xformers version: 0. apt install ffmpeg. ComfyUIが 圧倒的に早く （22%程度高速）、. Here’s a concise guide on how to interact with and manage nodes for an optimized user experience. Sort by: Search Comments. pip install xformers. Well it simply uses comfy's xformers check, which passes if it's installed and can be imported, and that it's not disabled by the command line argument --disable-xformers. You switched accounts on another tab or window. 4 cu124 with python 3. \p ython_embeded \p ython. Or --lowvram if you want it to use less. という感じ。. ちなみにこの 制作了中文版ComfyUI插件与节点汇总表，项目详见：【腾讯文档】ComfyUI 插件（模组）+ 节点（模块）汇总 【Zho】 20230916 近期谷歌Colab禁止了免费层运行SD，所以专门做了Kaggle平台的免费云部署，每周30小时免费冲浪时间，项目详见： Kaggle ComfyUI云部署1. Oct 18, 2023 · QualityOfLifeSuit_Omar92_DIR: I:\AI\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-QualityOfLifeSuit_Omar92 Total VRAM 4096 MB, total RAM 24468 MB Trying to enable lowvram mode because your GPU seems to have 4GB or less. py bdist_wheel. whl, change the name of the file in the command below if the name is different: . save_image: should GIF be saved to disk. Features. ## insatll xformers match your torch,for torch==2. "Queue prompt"をクリック。. If you have another Stable Diffusion UI you might be able to reuse the dependencies. There's these if you want it to use more vram: --gpu-only --highvram. Add a Comment. The current workaround is to disable xformers with --disable-xformers when booting ComfyUI. On my 3090 TI I get a 5-10% performance increase versus the old standalone. But for many nodes, most the more heavy CN preprocessors for exemple (geowizard, depthfm etc) and many other Xformers is mandatory, without it Jun 23, 2023 · What should be fixed The script should not force the installation of xformers 0. 3. 24 Set vram state to: NORMAL_VRAM Device: cuda:0 NVIDIA GeForce RTX 4090 : cudaMallocAsync VAE dtype: torch. 以下のサイトで公開されている Welcome to the unofficial ComfyUI subreddit. for Windows,you can install ffmpeg by WingetUI automatically. Combine GIF frames and produce the GIF image. 1 cu121. 22的一般能在1800×1800上快20-30秒，搞不懂为什么这么多人喜欢杠，显卡是笔记本3070ti，优化设置中等 Jun 20, 2023 · I started running in to the following issue on the latest when I launch with either. /venv/scripts Jul 30, 2023 · PyTorch limit (set by user-supplied memory fraction) : 17179869184. Thx! AnimateDiffCombine. Please scroll up your comfyUI console, it should tell you which package caused the import failure, also make sure to use the correct run_nvidia_gpu_miniconda. Reload to refresh your session. Set up a new domain A record that points to the Server IP Address. VRAM使用も少ない （半分程度）ことがわかる。. whl file to the base directory of stable-diffusion-webui. You signed out in another tab or window. 0, and we have also applied a patch to the pycocotools dependency for Windows environment in ddetailer. 1 prebuilt wheel. Welcome to the unofficial ComfyUI subreddit. Adding a Node: Simply right-click on any vacant space. But you can force it to do whatever you want by adding that into the command line. Asynchronous Queue system. So I downgraded torch to 2. Hopefully people can submit traces, screenshots to help us better understand why A1111 is slow. For example, comfyui. It's roughly the same speed as the previous version in ComfyUI inference but it at least allowed getting everything reasonably up to date without issues. to get started. I for some reason had to reinstall comfyui, and then this popped up again for me too, despite having xformers installed. After xFormers is installed, you can use enable_xformers_memory_efficient_attention() for faster inference and reduced memory consumption as shown in this section. d20230316 Set vram state to: HIGH_VRAM Device: cuda:0 NVIDIA GeForce RTX 4070 Ti Apr 10, 2023 · So far, we feel that working with it a slightly more overhead than working in Auto1111, but we have a lot more experience with the latter. onnx. 6) I tried upgrading Torch for Comfy and it crashed comfy completely. x Formers. post1+cu118) and it solved my problem. Mar 5, 2024 · JYLCode commented Mar 5, 2024. com. py --windows-standalone-build ** ComfyUI startup time: 2024-02-12 11:34:27. I'm using the portable version of ComfyUI and I started wondering should I update my xformers installation? When I started looking for answers on Reddit, I found a few posts saying that xformers is no longer necessary. 1935 64 bit (AMD64)] ** Python executable: D: \C omfyUI ComfyUI/ComfyUI - A powerful and modular stable diffusion GUI. 7 (you have 3. 0 、 Kaggle It's an xformers bug accidentally triggered by the way the original AnimateDiff CrossAttention is passed in. If --listen is provided without an argument, it defaults to 0. That's rather impressive. 04 with an RTX 3090. View full answer. Acess the server using SSH as a non-root sudo user. 但是，是的，xformers 安装起来经常有问题，可能不值得这么麻烦。 I am also worried that after this is fixed, other plug-ins will conflict or report errors, which will be a big failure for me. If I use the -h flag, listen is still showing as an option. Currently even if this can run without xformers, the memory usage is huge. Provide details and share your research! But avoid …. 前回の記事では、AI動画生成ツールのAnimateDiffと「ControlNet」を組み合わせることで、特定のモーションをアニメで再現しました。. Answered by comfyanonymous on Aug 8, 2023. Author. I'm using xformers on both systems and were testing with the same I'm running ComfyUI on ubuntu linux 22. 3. Feb 24, 2024 · ComfyUI is a node-based interface to use Stable Diffusion which was created by comfyanonymous in 2023. Typically, it is located in \ComfyUI_windows_portable\python Dec 29, 2023 · ComfyUI は、Stable Diffusion 用のノードベースのユーザー インターフェイスです。ここではWindowsのインストールを「安全に、完璧に」インストールする方法について説明します。 注意！ ComfyUIはローカル環境にインストールした後、使いたい拡張機能やモデルを別途、インストールする必要があり Jan 29, 2023 · 最後にcd で ComfyUI の中に入っていきます。 pip のアップデート. warnings. Installing ComfyUI. 最近ではSDXLモデルでの生成速度の早さ、消費VRAM量の少なさ（1304x768の生成時で6GB程度）から注目を浴びています。. Research first: xFormers contains bleeding-edge components, that are not yet available in mainstream libraries like PyTorch. 8 seconds: D:\ComfyUI\custom_nodes\ComfyUI-Manager Total VRAM 12287 MB, total RAM 65481 MB xformers version: 0. That's the default. 11 with pytorch 2. So, i'm stuck at using only 50% of the VRAM or nothing. The xformers issue should not pop up at all, as How to use. Install the ComfyUI dependencies. Here’s the link to the previous update in case you missed it. If you don't want this use: --normalvram xformers version: 0. ago. pip install xformers==0. BTW, I prefer not to. 0 seconds: E:\Blender_ComfyUI\ComfyUI\custom_nodes\ComfyUI-Manager. At the same time, we developed a few workflows that are just tailored to specific tasks (for example, testing different VAEs), and having the whole chain in front of us really helps us ensure that we are changing just one thing at a time for our experiments. Speed and memory usage of xFormers and Torch vary depending on the application scenario and require specific testing. Not Found. loop_count: use 0 for infinite loop. Sep 11, 2023 · You signed in with another tab or window. 0 seconds: X:\A1111\StabilityMatrix-win-x64\Data\Packages\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-Manager. Built with efficiency in mind: Because speed of iteration matters, components are as fast and memory-efficient as possible. 2+cu121) Python 3. 0+cu121. Asking for help, clarification, or responding to other answers. XFormers aims at being able to reproduce most architectures in the Transformer-family SOTA,defined as compatible and combined building blocks as opposed to monolithic models. Copy the download link from there and move to your stable diffusion directory. Jan 26, 2024 · Learn how to speed up your image generation with XFormers, a powerful tool that boosts Stable Diffusion performance. 强大且模块化的 stable diffusion 图形用户界面和后端设计. py --force-fp16. info command showed good versions of everything. A Deep Dive into ComfyUI Nodes. A platform for writing and freely expressing thoughts on various topics. bfloat16 Using Oct 19, 2023 · Hi, I have problem until yesterday. post4 Set vram state to: NORMAL_VRAM Device: cuda:0 NVIDIA GeForce RTX 3080 Ti : cudaMallocAsync VAE dtype: torch. Follow the steps to install XFormers for Automatic1111 or ComfyUI, and explore advanced settings and applications. Im using SDXL_1_0ComfyUI on google colab. To use video formats, you'll need ffmpeg installed and Sep 6, 2023 · ComfyUIでは、通常起動すると、ディフォルトでXformersがオンになっています。 これは画像生成速度を大幅に向上しつつ、VRAM消費量を抑えることができる機能で、対応するGPUであれば適応されます。 Oct 8, 2023 · Here’s what’s new recently in ComfyUI. py", line 380, in wrapped_function raise RuntimeError(f"An xformers bug was encountered in AnimateDiff - this is unexpected, \ Sep 11, 2023 · Steps to reproduce the behavior: i'm using the diffusers repo to export the unet to onnx model. If you do simple t2i or i2i you don't need xformers anymore, pytorch attention is enough. DWPose might run very slowly") Loaded ControlNetPreprocessors nodes from E:\ComfyUI\ComfyUI_windows_portable\ComfyUI\custom_nodes\comfyui_controlnet_aux. txt. I will start wit Sep 22, 2023 · ComfyUI AnimateDiffについての記事は、以下のnoteにまとめています。. xFormers was built for: PyTorch 2. bfloat16 Mar 14, 2024 · Thanks for contributing an answer to Stack Overflow! Please be sure to answer the question. *注意：このColabは、Google Colab Pro/Pro+で使用してください。無料版Colabでは画像生成AIの使用が規制されています。 Follow the ComfyUI manual installation instructions for Windows and Linux. This is the community-maintained repository of documentation related to ComfyUI open in new window, a powerful and modular stable diffusion GUI and backend. The red section contains parameters that can be adjusted according to your needs. May 7, 2024 · This thread will be used for performance profiling analysis for Forge/A1111/ComfyUI. 以下链接是一些 工作流示例 ，您可以通过这些示例了解到借助 ComfyUI 可以做什么：. To give you an idea of how powerful it is: StabilityAI, the creators of Stable Diffusion, use ComfyUI to test Stable Diffusion internally. I tried --gpu-only (OOM results) and --highvram. 0 but flash-attention 2 was difficult when I tried, somebody has since fixed it I think (and you'll need both to take full advantage) Xformers is a bigger advantage on larger generations. | ComfyUI 大号整合包，预装大量自定义节点（不含SD模型） - Releases · YanWenKun/ComfyUI-Windows-Portable Feb 10, 2024 · ldm is ComfyUI's modified version of CompVis ldm, and 2nd edited by @lllyasviel to accomondate our need, especially for AnimateDiff (for me, actually) modules is almost entirely from ComfyUI, with small modification for custom need. The text was updated successfully, but these errors were encountered: Dec 19, 2023 · ComfyUI was created in January 2023 by Comfyanonymous, who created the tool to learn how Stable Diffusion works. 17. The aim of this page is to get you up and running with ComfyUI, running your first gen, and providing some suggestions for the next steps to explore. I think I've heard that Xformers speeds up the image creation process, but that also brings some problems. 2, and re-installed xformers 0. See full list on github. apt update. frame_rate: number of frame per second. We’re on a journey to advance and democratize artificial intelligence through open source and open science. VAEは選択されていない模様. 04 A100 Vultr GPU Stack server using the Vultr marketplace application with at least 40 GB of GPU RAM. memory_efficient_attention( The text was updated successfully, but these errors were encountered: But yeah, xformers is often problematic to install, might not be worth the trouble. , NVidia) Install the rest with: pip install -r requirements. When the flow reaches the Stable Video Diffusion Sampler node, execution ceases. Aug 9, 2023 · Gourieff changed the title Can't start comfyui after following install instructions [SOLVED] Can't start comfyui after following install instructions Aug 11, 2023 Copy link Mapleshade20 commented Aug 26, 2023 Installing xFormers. Saved searches Use saved searches to filter your results more quickly Oct 4, 2023 · The build still completed successfully and the xformers. Delerium76. In our tests, the optimizations performed in the attention blocks allow for both faster speed and reduced memory consumption. This function is the most heavily used one, and controlnet. All those bucket seams need post processing to be able to apply different filter adjustments and composite them. Set cuda device to: 0 Total VRAM 12282 MB, total RAM 32694 MB xformers version: 0. The cross-attention mechanism in PyTorch 2. exe -m pip install -U xformers --no-dependencies) Dec 14, 2023 · ** ComfyUI start up time: 2023-12-14 16:46:05. py", line 249, in attention out = xformers. 此UI界面是基于 图形/节点/流程图 设计的，允许您设计和执行stable diffusion的任何流程。. Mar 12, 2023 · This unstable nightly pytorch build is for people who want to test latest pytorch to see if it gives them a performance boost. Nodes/graph/flowchart interface to experiment and create complex Stable Diffusion workflows without needing to code anything. py. Aug 12, 2023 · 概要. 12. In xformers directory, navigate to the dist folder and copy the . WARNING [XFORMERS]: xFormers can't load C++/CUDA extensions. I'm running that with the Pytorch 2. Belittling their efforts will get you banned. They have since hired Comfyanonymous to help them work on internal tools. I found out that after updating ComfyUI, it now uses torch 2. 2. /ComfyUI/main. A lot of people are just discovering this technology, and want to show off what they created. 0+cu121 pip install xformers xformers version: 0. ComfyUI 示例：https You signed in with another tab or window. Standalone Windows Package Updates. symbolic_registry. ComfyUI’s graph-based design is hinged on nodes, making them an integral aspect of its interface. 22. 最後のところに画像が生成されていればOK。. Total VRAM 4096 MB, total RAM 14188 MB Trying to enable lowvram mode because your GPU seems to have 4GB or less. D: \C omfyUI_windows_portable \n ew_ComfyUI_windows_portable_nvidia_cu121_or_cpu \C omfyUI_windows_portable >. when i enable the xformers, there will be this error: torch. 4. Many optimizations: Only re-executes the parts of the workflow that changes between executions. Might wanna check which pip it is using when you type in the command. pip install xformers (when a prebuild wheel is available, it take a few days in general after a pytorch update)/. post7. 10+CUDA12. 0 、 Kaggle latents = wrap_function_to_inject_xformers_bug_info(orig_comfy_sample)(model, noise, *args, **kwargs) ^^^^^ File "D:\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-AnimateDiff-Evolved\animatediff\utils_model. py", line 1, in You signed in with another tab or window. xFormers contains its own CUDA kernels, but dispatches to other libraries when relevant. 500. 0 seconds: D:\ComfyUI\custom_nodes\rgthree-comfy 18. If you are trying to export a custom operator Aug 7, 2023 · You signed in with another tab or window. py File "D:\Desktop\comfyui\ComfyUI. Apr 26, 2024 · This is the second time that has happened since yesterday. The standalone windows package now uses python 3. The current workaround is to disable xformers with "--disable-xformers" when booting ComfyUI. example. Note that this package contains nightly torch 2. tjedwards. 这是测试结果，这次用新版的xformers还跑的额外快，0. Unlike other Stable Diffusion tools that have basic text fields where you enter values and information for generating an image, a node-based interface is different in the sense that you’d have to create nodes to build a workflow to generate images. Work around: Shorter your prompt and negative prompt; Reduce resolution. 11. Recommended to use xformers if possible: Not by default. Apr 27, 2023 · ComfyUI. UnsupportedOperatorError: ONNX export failed on an operator with unrecognized namespace xformers_flash :: flash_fwd. 本記事では手動でインストールを行い、SDXLモデルで Apr 29, 2024 · Project description. ## in ComfyUI/custom_nodes. ← PyTorch 2. x and SDXL. exe -s ComfyUI\main. GIF split into multiple scenes. Project details. If you want to update to it you have to download a new version of the standalone. linux環境のほうが高速だと思っていたが、実測値を見る限りWSLにする意味はあまりないようだ。. 21 Set vram state to: LOW_VRAM out = xformers. ComfyUI checks what your hardware is and determines what is best. com Aug 2, 2023 · . XFormers: A collection of composable Transformer building blocks. txt; Each time there-after: Ensure that venv is active; python main. Feb 23, 2023 · A friend of mine for example is doing this on a GTX 960 (what a madman) and he's experiencing up to 3 times the speed when doing inference in ComfyUI over Automatic's. Prestartup times for custom nodes: 0. bat to start the comfyUI. . Jan 24, 2024 · File "C:\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-DragNUWA\dragnuwa\svd\modules\diffusionmodules\model. make sure ffmpeg is worked in your commandline for Linux. 3, it should automatically download xformers 0. 1). ext\Lib\site-packages\xformers\ops\fmha\dispatch. Note that --force-fp16 will only work if you installed the latest pytorch nightly. Sorry for my english its not perfect. py build. Sep 24, 2023 · 0. exe -s ComfyUI \m ain. 1. py --windows-standalone-build --xformers pause It appears you have installed xformers on your system's default Python but not in the virtual environment ComfyUI is using. memory_efficient_attention AttributeError: module 'xformers' has no attribute 'ops' The text was updated successfully, but these errors were encountered: This is a workflow that quickly Upscale images to 8K resolution; simply drag and drop your image and click to run. Dec 6, 2023 · ##### Prestartup times for custom nodes: 0. Starting from version 0. According to this issue , xFormers v0. I, on the other hand, are on a RTX 3090 TI and inference for me is 4 to 6 times slower than in Automatic's. python setup. 显卡型号没说喔！. 00 GiB. Bing-su/ dddetailer - The anime-face-detector used in ddetailer has been updated to be compatible with mmdet 3. Fully supports SD1. （ただしVRAMをわずかに抑えられる可能性はある）. Dec 31, 2023 · 結論. Oct 28, 2023 · 秋叶大佬整合包最新xformers真的太好用了. g. You shouldn't need xformers, as SDP is in all ways superior. (listens on all)`. Move and click to the address bar and type " cmd " to open command prompt on the respective directory. Models: Checkpoint: . Please share your tips, tricks, and workflows for using this software to create your AI art. exe -m pip install -r ComfyUI\custom_nodes\ComfyUI-DynamiCrafterWrapper\requirements. Switch between documentation themes. Oct 8, 2023 · Try removing the comfyui-animatediff folder (keeping only ComfyUI-AnimateDiff-Evolved, to avoid potential conflicts), then make sure AnimateDiff-Evolved is updated to the most recent version (you can do git pull in the AnimateDiff-Evolved folder just in case), and then attempt to run it again. SDXLの特徴の一つっぽいrefinerを使うには、それを使うようなフローを作る必要がある。. 6:8b6ee5b, Oct 2 2023, 14:57:12) [MSC v. まずは pip のアップデートをしておきましょう。(最新の xformers のため) pip install --upgrade pip pytorchインストール 今一度、自分がいる directory の場所が ComfyUI の下かどうかを確認しましょう。 In your confyui venv. 16 cannot be used for training (fine-tune or DreamBooth) in some GPUs. We recommend the use of xFormers for both inference and training. 2, torchvision to 0. では生成してみる。. Xformers has been removed. \python_embeded\python. This workflow is bulit on top of them and I learned a lot form their work. 0+cu121 with CUDA 1201 (you have 2. --listen [IP] Specify the IP address to listen on (default: 127. 16 which is compatible with torch 1. xj cn ly up lb rr dm vl yq an