Deprecated: The each() function is deprecated. This message will be suppressed on further calls in /home/zhenxiangba/zhenxiangba.com/public_html/phproxy-improved-master/index.php on line 456 sayakpaul (Sayak Paul)
https://huggingface.co/docs/diffusers/api/pipelines/consistency_models.\n\nUnconditional consistency model training: https://github.com/huggingface/diffusers/tree/main/examples/research_projects/consistency_training.","text":"Introduces a new framework for distillation by learning to map any point in a probability flow ordinary differential equation (ODE )to its origin on the trajectory. CMs also allow for 1-4 step sampling. \n\nPlay with them: https://huggingface.co/docs/diffusers/api/pipelines/consistency_models.\n\nUnconditional consistency model training: https://github.com/huggingface/diffusers/tree/main/examples/research_projects/consistency_training."},"id":"2303.01469","title":"Consistency Models","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2303.01469.png","upvotes":8,"publishedAt":"2023-03-02T18:30:16.000Z","isUpvotedByUser":false}],"position":2,"theme":"pink","private":false,"shareUrl":"https://hf.co/collections/sayakpaul/optimizing-diffusion-models","upvotes":21,"isUpvotedByUser":false}],"datasets":[{"author":"sayakpaul","downloads":20,"gated":false,"id":"sayakpaul/llm-bias-mt-bench","lastModified":"2025-12-23T06:56:24.000Z","datasetsServerInfo":{"viewer":"viewer","numRows":4764,"libraries":["datasets","mlcroissant"],"formats":["text"],"modalities":["text"]},"private":false,"repoType":"dataset","likes":0,"isLikedByUser":false,"isBenchmark":false},{"author":"sayakpaul","downloads":319,"gated":false,"id":"sayakpaul/good-meal","lastModified":"2025-11-08T06:58:07.000Z","datasetsServerInfo":{"viewer":"viewer","numRows":195,"libraries":["datasets","mlcroissant"],"formats":["imagefolder"],"modalities":["image"]},"private":false,"repoType":"dataset","likes":0,"isLikedByUser":false,"isBenchmark":false},{"author":"sayakpaul","downloads":10171,"gated":false,"id":"sayakpaul/sample-datasets","lastModified":"2025-08-11T04:50:10.000Z","datasetsServerInfo":{"viewer":"viewer","numRows":6,"libraries":["datasets","mlcroissant"],"formats":["imagefolder"],"modalities":["image"]},"private":false,"repoType":"dataset","likes":1,"isLikedByUser":false,"isBenchmark":false},{"author":"sayakpaul","downloads":2756,"gated":false,"id":"sayakpaul/butteflies_with_classes","lastModified":"2025-05-27T16:41:26.000Z","datasetsServerInfo":{"viewer":"preview","numRows":0,"libraries":[],"formats":[],"modalities":["image"]},"private":false,"repoType":"dataset","likes":0,"isLikedByUser":false,"isBenchmark":false},{"author":"sayakpaul","downloads":1038,"gated":false,"id":"sayakpaul/OmniEdit-mini","lastModified":"2025-01-05T11:28:09.000Z","datasetsServerInfo":{"viewer":"viewer","numRows":21064,"libraries":["datasets","dask","mlcroissant","polars"],"formats":["parquet"],"modalities":["image","text"]},"private":false,"repoType":"dataset","likes":5,"isLikedByUser":false,"isBenchmark":false},{"author":"sayakpaul","downloads":158,"gated":false,"id":"sayakpaul/video-dataset-disney-organized","lastModified":"2024-11-29T09:30:31.000Z","datasetsServerInfo":{"viewer":"viewer","numRows":69,"libraries":["datasets","mlcroissant"],"formats":[],"modalities":["text","video"]},"private":false,"repoType":"dataset","likes":6,"isLikedByUser":false,"isBenchmark":false},{"author":"sayakpaul","downloads":10,"gated":false,"id":"sayakpaul/pick-a-pic-v2-unique-prompts","lastModified":"2024-11-09T21:20:47.000Z","datasetsServerInfo":{"viewer":"viewer","numRows":58960,"libraries":["datasets","pandas","mlcroissant","polars"],"formats":["parquet"],"modalities":["text"]},"private":false,"repoType":"dataset","likes":1,"isLikedByUser":false,"isBenchmark":false},{"author":"sayakpaul","downloads":30,"gated":false,"id":"sayakpaul/poses-controlnet-dataset","lastModified":"2024-08-29T20:01:48.000Z","datasetsServerInfo":{"viewer":"viewer","numRows":496,"libraries":["datasets","pandas","mlcroissant","polars"],"formats":["parquet"],"modalities":["image","text"]},"private":false,"repoType":"dataset","likes":7,"isLikedByUser":false,"isBenchmark":false},{"author":"sayakpaul","downloads":55,"gated":false,"id":"sayakpaul/torchao-diffusers","lastModified":"2024-08-28T05:37:30.000Z","private":false,"repoType":"dataset","likes":0,"isLikedByUser":false,"isBenchmark":false},{"author":"sayakpaul","downloads":769,"gated":false,"id":"sayakpaul/pickapic_v2_webdataset","lastModified":"2024-04-04T09:45:58.000Z","datasetsServerInfo":{"viewer":"viewer-partial","numRows":8700,"libraries":["datasets","webdataset","mlcroissant"],"formats":["webdataset"],"modalities":["image","text"]},"private":false,"repoType":"dataset","likes":2,"isLikedByUser":false,"isBenchmark":false}],"models":[{"author":"sayakpaul","authorData":{"_id":"5f7fbd813e94f16a85448745","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1649681653581-5f7fbd813e94f16a85448745.jpeg","fullname":"Sayak Paul","name":"sayakpaul","type":"user","isPro":true,"isHf":true,"isHfAdmin":false,"isMod":false,"followerCount":852,"isUserFollowing":false},"downloads":0,"gated":false,"id":"sayakpaul/temp-new-model","availableInferenceProviders":[],"lastModified":"2025-12-24T13:53:33.000Z","likes":0,"private":false,"repoType":"model","isLikedByUser":false},{"author":"sayakpaul","authorData":{"_id":"5f7fbd813e94f16a85448745","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1649681653581-5f7fbd813e94f16a85448745.jpeg","fullname":"Sayak Paul","name":"sayakpaul","type":"user","isPro":true,"isHf":true,"isHfAdmin":false,"isMod":false,"followerCount":852,"isUserFollowing":false},"downloads":72,"gated":false,"id":"sayakpaul/different-lora-from-civitai","availableInferenceProviders":[],"lastModified":"2025-08-07T06:16:24.000Z","likes":1,"private":false,"repoType":"model","isLikedByUser":false,"numParameters":11901408320},{"author":"sayakpaul","authorData":{"_id":"5f7fbd813e94f16a85448745","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1649681653581-5f7fbd813e94f16a85448745.jpeg","fullname":"Sayak Paul","name":"sayakpaul","type":"user","isPro":true,"isHf":true,"isHfAdmin":false,"isMod":false,"followerCount":852,"isUserFollowing":false},"downloads":36,"gated":false,"id":"sayakpaul/flux-diffusers-gguf","availableInferenceProviders":[],"lastModified":"2025-06-10T04:12:16.000Z","likes":1,"private":false,"repoType":"model","isLikedByUser":false,"numParameters":11901408320},{"author":"sayakpaul","authorData":{"_id":"5f7fbd813e94f16a85448745","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1649681653581-5f7fbd813e94f16a85448745.jpeg","fullname":"Sayak Paul","name":"sayakpaul","type":"user","isPro":true,"isHf":true,"isHfAdmin":false,"isMod":false,"followerCount":852,"isUserFollowing":false},"downloads":15,"gated":false,"id":"sayakpaul/trained-lumina2-lora-yarn","availableInferenceProviders":[],"lastModified":"2025-02-20T06:08:21.000Z","likes":3,"pipeline_tag":"text-to-image","private":false,"repoType":"model","isLikedByUser":false,"widgetOutputUrls":["https://huggingface.co/sayakpaul/trained-lumina2-lora-yarn/resolve/main/yarn_lora.png","https://huggingface.co/sayakpaul/trained-lumina2-lora-yarn/resolve/main/yarn_lora_You_are_an_assistant_designed_to_generate_superior_images_with_a_dark_overall_theme.png","https://huggingface.co/sayakpaul/trained-lumina2-lora-yarn/resolve/main/yarn_lora_You_are_an_assistant_designed_to_generate_superior_images_with_a_bright_and_shiny_overall_.png"]},{"author":"sayakpaul","authorData":{"_id":"5f7fbd813e94f16a85448745","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1649681653581-5f7fbd813e94f16a85448745.jpeg","fullname":"Sayak Paul","name":"sayakpaul","type":"user","isPro":true,"isHf":true,"isHfAdmin":false,"isMod":false,"followerCount":852,"isUserFollowing":false},"downloads":62,"gated":false,"id":"sayakpaul/FLUX.1-dev-edit-v0","availableInferenceProviders":[],"lastModified":"2025-01-21T12:42:35.000Z","likes":46,"pipeline_tag":"text-to-image","private":false,"repoType":"model","isLikedByUser":false,"widgetOutputUrls":["https://huggingface.co/datasets/sayakpaul/sample-datasets/resolve/main/flux-edit-artifacts/edited_car.jpg","https://huggingface.co/datasets/sayakpaul/sample-datasets/resolve/main/flux-edit-artifacts/edited_green_creature.jpg","https://huggingface.co/datasets/sayakpaul/sample-datasets/resolve/main/flux-edit-artifacts/edited_mushroom.jpg","https://huggingface.co/datasets/sayakpaul/sample-datasets/resolve/main/flux-edit-artifacts/edited_norte_dam.jpg"]},{"author":"sayakpaul","authorData":{"_id":"5f7fbd813e94f16a85448745","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1649681653581-5f7fbd813e94f16a85448745.jpeg","fullname":"Sayak Paul","name":"sayakpaul","type":"user","isPro":true,"isHf":true,"isHfAdmin":false,"isMod":false,"followerCount":852,"isUserFollowing":false},"downloads":13,"gated":false,"id":"sayakpaul/cartoon-control-lr_1e-4-wd_1e-4-gs_10.0-cd_0.1","availableInferenceProviders":[],"lastModified":"2025-01-05T12:16:33.000Z","likes":6,"pipeline_tag":"text-to-image","private":false,"repoType":"model","isLikedByUser":false},{"author":"sayakpaul","authorData":{"_id":"5f7fbd813e94f16a85448745","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1649681653581-5f7fbd813e94f16a85448745.jpeg","fullname":"Sayak Paul","name":"sayakpaul","type":"user","isPro":true,"isHf":true,"isHfAdmin":false,"isMod":false,"followerCount":852,"isUserFollowing":false},"downloads":22,"gated":false,"id":"sayakpaul/q8-ltx-video","availableInferenceProviders":[],"lastModified":"2025-01-02T11:51:18.000Z","likes":7,"private":false,"repoType":"model","isLikedByUser":false},{"author":"sayakpaul","authorData":{"_id":"5f7fbd813e94f16a85448745","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1649681653581-5f7fbd813e94f16a85448745.jpeg","fullname":"Sayak Paul","name":"sayakpaul","type":"user","isPro":true,"isHf":true,"isHfAdmin":false,"isMod":false,"followerCount":852,"isUserFollowing":false},"downloads":13,"gated":false,"id":"sayakpaul/yarn_art_lora_sana","availableInferenceProviders":[],"lastModified":"2024-12-16T08:21:21.000Z","likes":1,"pipeline_tag":"text-to-image","private":false,"repoType":"model","isLikedByUser":false,"widgetOutputUrls":["https://huggingface.co/sayakpaul/yarn_art_lora_sana/resolve/main/image_0.png","https://huggingface.co/sayakpaul/yarn_art_lora_sana/resolve/main/image_1.png","https://huggingface.co/sayakpaul/yarn_art_lora_sana/resolve/main/image_2.png","https://huggingface.co/sayakpaul/yarn_art_lora_sana/resolve/main/image_3.png"]},{"author":"sayakpaul","authorData":{"_id":"5f7fbd813e94f16a85448745","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1649681653581-5f7fbd813e94f16a85448745.jpeg","fullname":"Sayak Paul","name":"sayakpaul","type":"user","isPro":true,"isHf":true,"isHfAdmin":false,"isMod":false,"followerCount":852,"isUserFollowing":false},"downloads":15,"gated":false,"id":"sayakpaul/bnb-single-file-checkpoint-from-civitai","availableInferenceProviders":[],"lastModified":"2024-12-04T02:00:36.000Z","likes":0,"private":false,"repoType":"model","isLikedByUser":false},{"author":"sayakpaul","authorData":{"_id":"5f7fbd813e94f16a85448745","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1649681653581-5f7fbd813e94f16a85448745.jpeg","fullname":"Sayak Paul","name":"sayakpaul","type":"user","isPro":true,"isHf":true,"isHfAdmin":false,"isMod":false,"followerCount":852,"isUserFollowing":false},"downloads":5,"gated":false,"id":"sayakpaul/mochi-lora-dissolve","availableInferenceProviders":[],"lastModified":"2024-11-29T20:05:02.000Z","likes":2,"pipeline_tag":"text-to-video","private":false,"repoType":"model","isLikedByUser":false,"widgetOutputUrls":["https://huggingface.co/sayakpaul/mochi-lora-dissolve/resolve/main/final_video_0.mp4"]}],"buckets":[],"numBuckets":0,"numberLikes":222,"papers":[{"id":"2510.05091","title":"Factuality Matters: When Image Generation and Editing Meet Structured\n Visuals","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2510.05091.png","upvotes":20,"publishedAt":"2025-10-06T17:56:55.000Z","isUpvotedByUser":false},{"id":"2505.10046","title":"Exploring the Deep Fusion of Large Language Models and Diffusion\n Transformers for Text-to-Image Synthesis","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2505.10046.png","upvotes":9,"publishedAt":"2025-05-15T07:43:23.000Z","isUpvotedByUser":false},{"id":"2504.16080","title":"From Reflection to Perfection: Scaling Inference-Time Optimization for\n Text-to-Image Diffusion Models via Reflection Tuning","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2504.16080.png","upvotes":15,"publishedAt":"2025-04-22T17:58:07.000Z","isUpvotedByUser":false},{"id":"2503.09641","title":"SANA-Sprint: One-Step Diffusion with Continuous-Time Consistency\n Distillation","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2503.09641.png","upvotes":42,"publishedAt":"2025-03-12T04:53:07.000Z","isUpvotedByUser":false},{"id":"2412.03895","title":"A Noise is Worth Diffusion Guidance","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2412.03895.png","upvotes":29,"publishedAt":"2024-12-05T06:09:56.000Z","isUpvotedByUser":false},{"id":"2412.01487","title":"FastRM: An efficient and automatic explainability framework for\n multimodal generative models","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2412.01487.png","upvotes":1,"publishedAt":"2024-12-02T13:39:29.000Z","isUpvotedByUser":false},{"id":"2408.13467","title":"LlamaDuo: LLMOps Pipeline for Seamless Migration from Service LLMs to\n Small-Scale Local LLMs","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2408.13467.png","upvotes":25,"publishedAt":"2024-08-24T05:03:08.000Z","isUpvotedByUser":false},{"id":"2406.06424","title":"Margin-aware Preference Optimization for Aligning Diffusion Models\n without Reference","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2406.06424.png","upvotes":15,"publishedAt":"2024-06-10T16:14:45.000Z","isUpvotedByUser":false},{"id":"2404.01197","title":"Getting it Right: Improving Spatial Consistency in Text-to-Image Models","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2404.01197.png","upvotes":31,"publishedAt":"2024-04-01T15:55:25.000Z","isUpvotedByUser":false},{"id":"2402.17412","title":"DiffuseKronA: A Parameter Efficient Fine-tuning Method for Personalized\n Diffusion Model","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2402.17412.png","upvotes":23,"publishedAt":"2024-02-27T11:05:34.000Z","isUpvotedByUser":false},{"id":"2401.05252","title":"PIXART-δ: Fast and Controllable Image Generation with Latent\n Consistency Models","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2401.05252.png","upvotes":49,"publishedAt":"2024-01-10T16:27:38.000Z","isUpvotedByUser":false},{"id":"2401.02677","title":"Progressive Knowledge Distillation Of Stable Diffusion XL Using Layer\n Level Loss","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2401.02677.png","upvotes":25,"publishedAt":"2024-01-05T07:21:46.000Z","isUpvotedByUser":false},{"id":"2107.08369","title":"Flood Segmentation on Sentinel-1 SAR Imagery with Semi-Supervised\n Learning","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2107.08369.png","upvotes":0,"publishedAt":"2021-07-18T05:42:10.000Z","isUpvotedByUser":false},{"id":"2105.08819","title":"Fast and Accurate Quantized Camera Scene Detection on Smartphones,\n Mobile AI 2021 Challenge: Report","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2105.08819.png","upvotes":1,"publishedAt":"2021-05-17T13:55:38.000Z","isUpvotedByUser":false},{"id":"2105.07581","title":"Vision Transformers are Robust Learners","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2105.07581.png","upvotes":1,"publishedAt":"2021-05-17T02:39:22.000Z","isUpvotedByUser":false},{"id":"2009.12007","title":"G-SimCLR : Self-Supervised Contrastive Learning with Guided Projection\n via Pseudo Labelling","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2009.12007.png","upvotes":0,"publishedAt":"2020-09-25T02:25:37.000Z","isUpvotedByUser":false},{"id":"1905.13294","title":"A Review of Deep Learning with Special Emphasis on Architectures,\n Applications and Recent Trends","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/1905.13294.png","upvotes":0,"publishedAt":"2019-05-30T20:38:50.000Z","isUpvotedByUser":false}],"posts":[{"slug":"568017454065017","content":[{"type":"text","value":"Fast LoRA inference for Flux with Diffusers and PEFT 🚨","raw":"Fast LoRA inference for Flux with Diffusers and PEFT 🚨"},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"There are great materials that demonstrate how to optimize inference for popular image generation models, such as Flux. However, very few cover how to serve LoRAs fast, despite LoRAs being an inseparable part of their adoption.","raw":"There are great materials that demonstrate how to optimize inference for popular image generation models, such as Flux. However, very few cover how to serve LoRAs fast, despite LoRAs being an inseparable part of their adoption."},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"In our latest post, ","raw":"In our latest post, "},{"type":"mention","user":"BenjaminB","raw":"@BenjaminB"},{"type":"text","value":" and I show different techniques to optimize LoRA inference for the Flux family of models for image generation. Our recipe includes the use of:","raw":" and I show different techniques to optimize LoRA inference for the Flux family of models for image generation. Our recipe includes the use of:"},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"1. ","raw":"1. "},{"type":"inline_code","code":"torch.compile","raw":"`torch.compile`"},{"type":"new_line","raw":"\n"},{"type":"text","value":"2. Flash Attention 3 (when compatible)","raw":"2. Flash Attention 3 (when compatible)"},{"type":"new_line","raw":"\n"},{"type":"text","value":"3. Dynamic FP8 weight quantization (when compatible)","raw":"3. Dynamic FP8 weight quantization (when compatible)"},{"type":"new_line","raw":"\n"},{"type":"text","value":"4. Hotswapping for avoiding recompilation during swapping new LoRAs 🤯","raw":"4. Hotswapping for avoiding recompilation during swapping new LoRAs 🤯"},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"We have tested our recipe with Flux.1-Dev on both H100 and RTX 4090. We achieve at least a *2x speedup* in either of the GPUs. We believe our recipe is grounded in the reality of how LoRA-based use cases are generally served. So, we hope this will be beneficial to the community 🤗","raw":"We have tested our recipe with Flux.1-Dev on both H100 and RTX 4090. We achieve at least a *2x speedup* in either of the GPUs. We believe our recipe is grounded in the reality of how LoRA-based use cases are generally served. So, we hope this will be beneficial to the community 🤗"},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"Even though our recipe was tested primarily with NVIDIA GPUs, it should also work with AMD GPUs.","raw":"Even though our recipe was tested primarily with NVIDIA GPUs, it should also work with AMD GPUs."},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"Learn the details and the full code here:","raw":"Learn the details and the full code here:"},{"type":"new_line","raw":"\n"},{"type":"link","href":"https://huggingface.co/blog/lora-fast","raw":"https://huggingface.co/blog/lora-fast"}],"rawContent":"Fast LoRA inference for Flux with Diffusers and PEFT 🚨\n\nThere are great materials that demonstrate how to optimize inference for popular image generation models, such as Flux. However, very few cover how to serve LoRAs fast, despite LoRAs being an inseparable part of their adoption.\n\nIn our latest post, @BenjaminB and I show different techniques to optimize LoRA inference for the Flux family of models for image generation. Our recipe includes the use of:\n\n1. `torch.compile`\n2. Flash Attention 3 (when compatible)\n3. Dynamic FP8 weight quantization (when compatible)\n4. Hotswapping for avoiding recompilation during swapping new LoRAs 🤯\n\nWe have tested our recipe with Flux.1-Dev on both H100 and RTX 4090. We achieve at least a *2x speedup* in either of the GPUs. We believe our recipe is grounded in the reality of how LoRA-based use cases are generally served. So, we hope this will be beneficial to the community 🤗\n\nEven though our recipe was tested primarily with NVIDIA GPUs, it should also work with AMD GPUs.\n\nLearn the details and the full code here:\nhttps://huggingface.co/blog/lora-fast","author":{"_id":"5f7fbd813e94f16a85448745","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1649681653581-5f7fbd813e94f16a85448745.jpeg","fullname":"Sayak Paul","name":"sayakpaul","type":"user","isPro":true,"isHf":true,"isHfAdmin":false,"isMod":false,"followerCount":852,"isUserFollowing":false},"attachments":[{"type":"image","url":"https://cdn-uploads.huggingface.co/production/uploads/5f7fbd813e94f16a85448745/XVQTdsVKJZSxchsBS9EOv.png"}],"mentions":[{"_id":"62bf03d1e80cec527083cd66","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1656685953025-62bf03d1e80cec527083cd66.jpeg","fullname":"Benjamin Bossan","name":"BenjaminB","type":"user","isPro":false,"isHf":true,"isHfAdmin":false,"isMod":false,"followerCount":95,"isUserFollowing":false}],"reactions":[{"reaction":"🤗","users":["John6666","9voltfan2009","ethix"],"count":3},{"reaction":"🔥","users":["John6666","9voltfan2009","tolgacangoz"],"count":3}],"publishedAt":"2025-07-24T15:22:28.000Z","updatedAt":"2025-12-16T03:35:13.234Z","commentators":[{"_id":"68f7f0987acf8d6817fca61e","avatarUrl":"/avatars/14c91aa75e9e0c9ad58e00725973714f.svg","fullname":"Charles hugie","name":"bigpappic","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1,"isUserFollowing":false}],"url":"/posts/sayakpaul/568017454065017","totalUniqueImpressions":2346,"identifiedLanguage":{"language":"en","probability":0.9315197467803955},"numComments":3},{"slug":"442543668382056","content":[{"type":"text","value":"Diffusers supports a good variety of quantization backends. It can be challenging to navigate through them, given the complex nature of diffusion pipelines in general. ","raw":"Diffusers supports a good variety of quantization backends. It can be challenging to navigate through them, given the complex nature of diffusion pipelines in general. "},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"So, ","raw":"So, "},{"type":"mention","user":"derekl35","raw":"@derekl35"},{"type":"text","value":" set out to write a comprehensive guide that puts users in the front seat. Explore the different backends we support, learn the trade-offs they offer, and finally, check out the cool space we built that lets you compare quantization results.","raw":" set out to write a comprehensive guide that puts users in the front seat. Explore the different backends we support, learn the trade-offs they offer, and finally, check out the cool space we built that lets you compare quantization results."},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"Give it a go here:","raw":"Give it a go here:"},{"type":"new_line","raw":"\n"},{"type":"link","href":"https://lnkd.in/gf8Pi4-2","raw":"https://lnkd.in/gf8Pi4-2"}],"rawContent":"Diffusers supports a good variety of quantization backends. It can be challenging to navigate through them, given the complex nature of diffusion pipelines in general. \n\nSo, @derekl35 set out to write a comprehensive guide that puts users in the front seat. Explore the different backends we support, learn the trade-offs they offer, and finally, check out the cool space we built that lets you compare quantization results.\n\nGive it a go here:\nhttps://lnkd.in/gf8Pi4-2","author":{"_id":"5f7fbd813e94f16a85448745","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1649681653581-5f7fbd813e94f16a85448745.jpeg","fullname":"Sayak Paul","name":"sayakpaul","type":"user","isPro":true,"isHf":true,"isHfAdmin":false,"isMod":false,"followerCount":852,"isUserFollowing":false},"attachments":[{"type":"video","url":"https://cdn-uploads.huggingface.co/production/uploads/5f7fbd813e94f16a85448745/1pQXlnlzVA-OZ0LEcnCcQ.mp4"}],"mentions":[{"_id":"67c5c652f490a32d32b905ed","avatarUrl":"/avatars/e5b8331c9a96cd96b679f38afd30422e.svg","fullname":"Derek Liu","name":"derekl35","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":45,"isUserFollowing":false}],"reactions":[{"reaction":"🚀","users":["John6666","davidberenstein1957","KvLove","marcsun13","tolgacangoz"],"count":5},{"reaction":"🤗","users":["John6666","linoyts","davidberenstein1957","marcsun13"],"count":4}],"publishedAt":"2025-05-22T05:41:08.000Z","updatedAt":"2025-10-25T18:35:22.776Z","commentators":[{"_id":"603bdba23249b99991dbcbc4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/603bdba23249b99991dbcbc4/cxCnN1H-RXOhojHY3Wcxo.jpeg","fullname":"Tolga Cangöz","name":"tolgacangoz","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":29,"isUserFollowing":false},{"_id":"5f7fbd813e94f16a85448745","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1649681653581-5f7fbd813e94f16a85448745.jpeg","fullname":"Sayak Paul","name":"sayakpaul","type":"user","isPro":true,"isHf":true,"isHfAdmin":false,"isMod":false,"followerCount":852,"isUserFollowing":false}],"url":"/posts/sayakpaul/442543668382056","totalUniqueImpressions":2947,"identifiedLanguage":{"language":"en","probability":0.8442586660385132},"numComments":2}],"totalPosts":24,"spaces":[{"author":"sayakpaul","authorData":{"_id":"5f7fbd813e94f16a85448745","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1649681653581-5f7fbd813e94f16a85448745.jpeg","fullname":"Sayak Paul","name":"sayakpaul","type":"user","isPro":true,"isHf":true,"isHfAdmin":false,"isMod":false,"followerCount":852,"isUserFollowing":false},"colorFrom":"purple","colorTo":"indigo","createdAt":"2026-01-20T11:41:47.000Z","emoji":"🐢","id":"sayakpaul/github-repocard-creator","lastModified":"2026-01-20T12:02:21.000Z","likes":1,"pinned":false,"private":false,"sdk":"gradio","repoType":"space","runtime":{"stage":"SLEEPING","hardware":{"current":null,"requested":"cpu-basic"},"storage":null,"gcTimeout":172800,"replicas":{"requested":1},"devMode":false,"domains":[{"domain":"sayakpaul-github-repocard-creator.hf.space","stage":"READY"}]},"shortDescription":"Generate simple and clean GitHub repository cards.","title":"Github Repocard Creator","isLikedByUser":false,"trendingScore":0,"tags":["gradio","region:us"],"featured":false},{"author":"sayakpaul","authorData":{"_id":"5f7fbd813e94f16a85448745","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1649681653581-5f7fbd813e94f16a85448745.jpeg","fullname":"Sayak Paul","name":"sayakpaul","type":"user","isPro":true,"isHf":true,"isHfAdmin":false,"isMod":false,"followerCount":852,"isUserFollowing":false},"colorFrom":"gray","colorTo":"gray","createdAt":"2025-12-12T10:32:26.000Z","emoji":"💻","id":"sayakpaul/nb-fashion-moodboard","lastModified":"2025-12-15T04:43:04.000Z","likes":0,"pinned":false,"private":false,"sdk":"docker","repoType":"space","runtime":{"stage":"RUNNING","hardware":{"current":"cpu-basic","requested":"cpu-basic"},"storage":null,"gcTimeout":172800,"replicas":{"current":1,"requested":1},"devMode":false,"domains":[{"domain":"sayakpaul-nb-fashion-moodboard.hf.space","stage":"READY"}],"sha":"0465968f9243d847db586b6b9b4e0252c76c87af"},"shortDescription":"Nano Banana powered fashion moodboard","title":"NB Fashion Moodboard","isLikedByUser":false,"ai_short_description":"Generate a fashion moodboard from your ideas","ai_category":"Image Generation","trendingScore":0,"tags":["docker","region:us"],"featured":false},{"author":"sayakpaul","authorData":{"_id":"5f7fbd813e94f16a85448745","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1649681653581-5f7fbd813e94f16a85448745.jpeg","fullname":"Sayak Paul","name":"sayakpaul","type":"user","isPro":true,"isHf":true,"isHfAdmin":false,"isMod":false,"followerCount":852,"isUserFollowing":false},"colorFrom":"red","colorTo":"blue","createdAt":"2025-03-26T16:03:12.000Z","emoji":"🐠","id":"sayakpaul/civitai-to-hub","lastModified":"2025-03-26T16:37:52.000Z","likes":5,"pinned":false,"private":false,"sdk":"gradio","repoType":"space","runtime":{"stage":"SLEEPING","hardware":{"current":null,"requested":"cpu-basic"},"storage":null,"gcTimeout":172800,"replicas":{"requested":1},"devMode":false,"domains":[{"domain":"sayakpaul-civitai-to-hub.hf.space","stage":"READY"}]},"shortDescription":"Upload checkpoints from CivitAI to Hugging Face Hub.","title":"Civitai To Hub","isLikedByUser":false,"ai_short_description":"Upload CivitAI checkpoints to Hugging Face Hub","ai_category":"Other","trendingScore":0,"tags":["gradio","region:us"],"featured":false},{"author":"sayakpaul","authorData":{"_id":"5f7fbd813e94f16a85448745","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1649681653581-5f7fbd813e94f16a85448745.jpeg","fullname":"Sayak Paul","name":"sayakpaul","type":"user","isPro":true,"isHf":true,"isHfAdmin":false,"isMod":false,"followerCount":852,"isUserFollowing":false},"colorFrom":"yellow","colorTo":"purple","createdAt":"2025-02-10T05:22:20.000Z","emoji":"🚀","id":"sayakpaul/grade_images_with_gemini","lastModified":"2025-02-10T06:36:07.000Z","likes":9,"pinned":false,"private":false,"sdk":"gradio","repoType":"space","runtime":{"stage":"RUNNING","hardware":{"current":"cpu-basic","requested":"cpu-basic"},"storage":null,"gcTimeout":172800,"replicas":{"current":1,"requested":1},"devMode":false,"domains":[{"domain":"sayakpaul-grade-images-with-gemini.hf.space","stage":"READY"}],"sha":"137f70281b6de9f3d68d814a39e1e8841efb6dd5"},"shortDescription":"Uses Gemini 2.0 Flash to grade images.","title":"Grade Images with Gemini","isLikedByUser":false,"ai_short_description":"Grade images based on prompts","ai_category":"Image","trendingScore":0,"tags":["gradio","region:us"],"featured":false},{"author":"sayakpaul","authorData":{"_id":"5f7fbd813e94f16a85448745","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1649681653581-5f7fbd813e94f16a85448745.jpeg","fullname":"Sayak Paul","name":"sayakpaul","type":"user","isPro":true,"isHf":true,"isHfAdmin":false,"isMod":false,"followerCount":852,"isUserFollowing":false},"colorFrom":"gray","colorTo":"indigo","createdAt":"2025-01-16T03:16:44.000Z","emoji":"🐨","id":"sayakpaul/auto-release-notes-diffusers","lastModified":"2025-01-16T03:24:02.000Z","likes":0,"pinned":false,"private":false,"sdk":"gradio","repoType":"space","runtime":{"stage":"SLEEPING","hardware":{"current":null,"requested":"cpu-basic"},"storage":null,"gcTimeout":172800,"replicas":{"requested":1},"devMode":false,"domains":[{"domain":"sayakpaul-auto-release-notes-diffusers.hf.space","stage":"READY"}]},"title":"Github Release Notes for Diffusers","isLikedByUser":false,"ai_short_description":"Generate release notes for Hugging Face Diffusers","ai_category":"Text Generation","trendingScore":0,"tags":["gradio","region:us"],"featured":false},{"author":"sayakpaul","authorData":{"_id":"5f7fbd813e94f16a85448745","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1649681653581-5f7fbd813e94f16a85448745.jpeg","fullname":"Sayak Paul","name":"sayakpaul","type":"user","isPro":true,"isHf":true,"isHfAdmin":false,"isMod":false,"followerCount":852,"isUserFollowing":false},"colorFrom":"indigo","colorTo":"indigo","createdAt":"2022-12-21T12:11:30.000Z","emoji":"📈","id":"sayakpaul/demo-docker-gradio","lastModified":"2024-03-20T01:29:01.000Z","likes":22,"pinned":false,"private":false,"sdk":"docker","repoType":"space","runtime":{"stage":"RUNTIME_ERROR","hardware":{"current":null,"requested":"cpu-basic"},"storage":null,"gcTimeout":86400,"errorMessage":"Launch timed out, workload was not healthy after 30 min","replicas":{"requested":1},"devMode":false,"domains":[{"domain":"sayakpaul-demo-docker-gradio.hf.space","stage":"READY"}]},"title":"Demo Docker Gradio","isLikedByUser":false,"ai_short_description":"Upload an image to get a label","ai_category":"Image","trendingScore":0,"tags":["docker","region:us"],"featured":false}],"u":{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1649681653581-5f7fbd813e94f16a85448745.jpeg","isPro":true,"fullname":"Sayak Paul","user":"sayakpaul","orgs":[{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1583856921041-5dd96eb166059660ed1ee413.png","fullname":"Hugging Face","name":"huggingface","userRole":"write","type":"org","isHf":true,"details":"The AI community building the future.","plan":"team"},{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1668178606778-602e6dee60e3dd96631c906e.jpeg","fullname":"🧨Diffusers","name":"diffusers","userRole":"admin","type":"org","isHf":false,"details":"Solving everything with diffusion models!","plan":"team"},{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1676518945578-5f7fbd813e94f16a85448745.png","fullname":"TensorFlow TPU","name":"tf-tpu","userRole":"admin","type":"org","isHf":false},{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1653062536500-5e9ecfc04957053f60648a3e.png","fullname":"Hugging Face Internal Testing Organization","name":"hf-internal-testing","userRole":"admin","type":"org","isHf":false},{"avatarUrl":"https://www.gravatar.com/avatar/24fcdab3af50907ff0c3c7b6107b42f8?d=retro&size=100","fullname":"All Things ViTs","name":"all-things-vits","userRole":"admin","type":"org","isHf":false,"details":"Hosts materials for CVPR 2023 tutorial: All Things ViTs: Understanding and Interpreting Attention in Vision."},{"avatarUrl":"https://www.gravatar.com/avatar/ce7578bbeab5cc46cc9f6a3fdfa56453?d=retro&size=100","fullname":"Probing ViTs","name":"probing-vits","userRole":"admin","type":"org","isHf":false,"details":"We are interested to study the representations learned by Vision Transformers."},{"avatarUrl":"https://www.gravatar.com/avatar/774c7a66a86daf1586ba46ed68ed814a?d=retro&size=100","fullname":"Instruction-tuned Diffusion Models","name":"instruction-tuning-sd","userRole":"admin","type":"org","isHf":false,"details":"Instruction tuning, Diffusion models"},{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1654000604623-5dfcb1aada6d0311fd3d5448.png","fullname":"(De)fusing","name":"fusing","userRole":"admin","type":"org","isHf":false,"details":"Fusing diffusion models"},{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1655315569655-6064e095abd8d3692e3e2ed6.png","fullname":"Huggingface Projects","name":"huggingface-projects","userRole":"write","type":"org","isHf":false},{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/5f0c746619cb630495b814fd/j26aNEdiOgptZxJ6akGCC.png","fullname":"Hugging Face H4","name":"HuggingFaceH4","userRole":"write","type":"org","isHf":false,"details":"Aligning LLMs to be helpful, honest, harmless, and huggy (H4)","plan":"team"},{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1675366649660-6141a88b3a0ec78603c9e784.png","fullname":"Keras Dreambooth Event","name":"keras-dreambooth","userRole":"write","type":"org","isHf":false},{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1664215365576-5e3aec01f55e2b62848a5217.png","fullname":"Hugging Face OSS Metrics","name":"open-source-metrics","userRole":"write","type":"org","isHf":false},{"avatarUrl":"https://www.gravatar.com/avatar/1d4cc1ea6b1b75decd2804bf34ad6fe8?d=retro&size=100","fullname":"Deploy HF TF ViTs","name":"deploy-hf-tf-vit","userRole":"admin","type":"org","isHf":false},{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60a551a34ecc5d054c8ad93e/Ku5nM2bKq-8ZF3Jid1ocw.png","fullname":"Blog-explorers","name":"blog-explorers","userRole":"read","type":"org","isHf":false},{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60d3b57ad7b174177faabd6e/dQ07CG-GnKmcUnQhJ0NyT.png","fullname":"Personal Coding Assistant","name":"coding-assistant-custom","userRole":"admin","type":"org","isHf":false,"details":"Code language models."},{"avatarUrl":"https://www.gravatar.com/avatar/ef4ee7a58d260d39a69a6564a29496a1?d=retro&size=100","fullname":"Diffusers Demo at ICCV 2023","name":"iccv23-diffusers-demo","userRole":"admin","type":"org","isHf":false,"details":"Diffusion models"},{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/5f17f0a0925b9863e28ad517/exBCcWH8McHg1hVQAtvN5.png","fullname":"ZeroGPU Explorers","name":"zero-gpu-explorers","userRole":"admin","type":"org","isHf":false},{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6320c537a023aad6a7680c8b/ccyONqDXP6eOBiPPRKEGJ.png","fullname":"SPRIGHT","name":"SPRIGHT-T2I","userRole":"admin","type":"org","isHf":false,"details":"Diffusion models"},{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/5f5006390bcb0e71543cec17/MS-FNCjJ1CycH0feIYTzL.png","fullname":"PEFT","name":"PEFT","userRole":"admin","type":"org","isHf":false,"details":"🤗 parameter-efficient fine-tuning"},{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/626dc5105f7327906f0b2a4e/Kn-QtZjE6TJE-syTndXIW.jpeg","fullname":"VISIONx @ NYU","name":"nyu-visionx","userRole":"contributor","type":"org","isHf":false},{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/NuDI1YvNy95sOdRnsi3dT.png","fullname":"MaPO","name":"mapo-t2i","userRole":"admin","type":"org","isHf":false,"details":"Diffusion models, alignment"},{"avatarUrl":"https://www.gravatar.com/avatar/526387de98bd7d5b70e7b142c5bb63bf?d=retro&size=100","fullname":"diffusers-internal-dev","name":"diffusers-internal-dev","userRole":"admin","type":"org","isHf":false,"plan":"team"},{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6380ebb8471a4550ff255c62/FVURHcQfjRnHy4rkVRpQy.jpeg","fullname":"AuraFlow","name":"AuraDiffusion","userRole":"write","type":"org","isHf":false,"details":"Research on next generation open source image models"},{"avatarUrl":"https://www.gravatar.com/avatar/2b53f98c9971b411112f25d8af68b3a2?d=retro&size=100","fullname":"lawrence","name":"JunsongChen","userRole":"write","type":"org","isHf":false},{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63bbb4d09726f7e58f96b70b/BouvZzJeaEPeVxYiQxcw6.png","fullname":"Diffusion Guidance","name":"diffusion-guidance-ku-hf","userRole":"write","type":"org","isHf":false},{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/5dd96eb166059660ed1ee413/wdv3IKpGrD0uSvefLrYGN.png","fullname":"DDUF","name":"DDUF","userRole":"write","type":"org","isHf":false},{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6452d5ba3f80ad88c77b2f05/0J-xey5Z1dh9ZOyTyyTge.png","fullname":"kernels-community","name":"kernels-community","userRole":"contributor","type":"org","isHf":false,"plan":"team"},{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/631f58935ba8c026340b377c/NRiWDTlxfCWz1lz-Bz8jF.png","fullname":"LTX Collaborations","name":"LTX-Colab","userRole":"read","type":"org","isHf":false},{"avatarUrl":"https://www.gravatar.com/avatar/c98615f99e27d862af2babbd9fbaa72c?d=retro&size=100","fullname":"Hunyuan Community","name":"hunyuanvideo-community","userRole":"admin","type":"org","isHf":false},{"avatarUrl":"https://www.gravatar.com/avatar/fba1c9e380af2ea61baadc1c1e17761f?d=retro&size=100","fullname":"Finetrainers","name":"finetrainers","userRole":"admin","type":"org","isHf":false,"details":"Accessible and delightful video model training and fine-tuning"},{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/5f7fbd813e94f16a85448745/u7-k61BeAabAW9S4AUm94.png","fullname":"Diffusion CoT","name":"diffusion-cot","userRole":"admin","type":"org","isHf":false,"details":"diffusion","plan":"team"},{"avatarUrl":"https://www.gravatar.com/avatar/da380b44559b27fb213c0269c20b6a9a?d=retro&size=100","fullname":"Kernels Tests","name":"kernels-test","userRole":"admin","type":"org","isHf":false},{"avatarUrl":"https://www.gravatar.com/avatar/8c416c0eed04943e4dfd6ec7b6bf7340?d=retro&size=100","fullname":"Cinematic T2V","name":"CinematicT2vData","userRole":"admin","type":"org","isHf":false},{"avatarUrl":"https://www.gravatar.com/avatar/03d02b7289a47f027da28527841b59a0?d=retro&size=100","fullname":"Diffusers Internal Demos","name":"diffusers-internal-demos","userRole":"write","type":"org","isHf":false},{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6069de23a0e75b0dd0135620/8N3_y8YAgDrkhjO2sfvNd.png","fullname":"ZeroGPU AoTI","name":"zerogpu-aoti","userRole":"admin","type":"org","isHf":false,"details":"AoT compilation, ZeroGPU inference optimization"},{"avatarUrl":"https://www.gravatar.com/avatar/90b533ab1eba8fe26a90fab8fdc8addc?d=retro&size=100","fullname":"PyTorch Day India HF","name":"pt-day-ind-hf","userRole":"admin","type":"org","isHf":false}],"signup":{"github":"sayakpaul","homepage":"https://sayak.dev","twitter":"RisingSayak","details":"Diffusion models, representation learning"},"isHf":true,"isMod":false,"type":"user","theme":"dark","status":"training","primaryOrg":{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1583856921041-5dd96eb166059660ed1ee413.png","fullname":"Hugging Face","name":"huggingface","userRole":"write","type":"org","isHf":true,"details":"The AI community building the future.","plan":"team"}},"upvotes":81,"numFollowers":852,"numFollowingUsers":0,"numFollowingOrgs":61,"numModels":40,"numDatasets":24,"numSpaces":15,"isFollowing":false,"isFollower":false,"sampleFollowers":[{"user":"cwhuh","fullname":"Chaewon Huh","type":"user","_id":"661f95dcf1c30c7d59f2ec30","isPro":false,"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/ddWw8gGgEUmIHCKM3MjLS.jpeg"},{"user":"shivance","fullname":"Anshuman Mishra","type":"user","_id":"63657b0898da81987e23b15b","isPro":false,"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63657b0898da81987e23b15b/o-Dd615dxU_IAZwmj54h5.jpeg"},{"user":"EURO-KOLYAN","fullname":"Nikolay","type":"user","_id":"647f80d4d26579210f5cf4a2","isPro":false,"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/647f80d4d26579210f5cf4a2/9tMvcIelLUKQEhLSs3-h2.jpeg"},{"user":"Soony123","fullname":"sony","type":"user","_id":"65573b020f4493529792186c","isPro":false,"avatarUrl":"/avatars/32d8cc8b2f0187f828d8bc281747aa2f.svg"}],"isWatching":false,"hardwareItems":[{"sku":["Apple Silicon","-","Apple M2 Max"],"mem":32,"num":1}],"isIgnored":false,"acceptLanguages":["*"],"filters":{},"currentRepoPage":0}">
Fast LoRA inference for Flux with Diffusers and PEFT 🚨
There are great materials that demonstrate how to optimize inference for popular image generation models, such as Flux. However, very few cover how to serve LoRAs fast, despite LoRAs being an inseparable part of their adoption.
In our latest post, @BenjaminB and I show different techniques to optimize LoRA inference for the Flux family of models for image generation. Our recipe includes the use of:
1. torch.compile 2. Flash Attention 3 (when compatible) 3. Dynamic FP8 weight quantization (when compatible) 4. Hotswapping for avoiding recompilation during swapping new LoRAs 🤯
We have tested our recipe with Flux.1-Dev on both H100 and RTX 4090. We achieve at least a *2x speedup* in either of the GPUs. We believe our recipe is grounded in the reality of how LoRA-based use cases are generally served. So, we hope this will be beneficial to the community 🤗
Even though our recipe was tested primarily with NVIDIA GPUs, it should also work with AMD GPUs.