If you haven't already, you can install the Transformers.js JavaScript library from NPM using
SAVRN Model Hub
Open-Weight Models
An open-weight model is an AI model whose trained weights are published for anyone to download. The weights are what the model learned in training. With a copy of them you can run the model on hardware you control and train it further on your own data.
Open weights are not the same as open source. Many publishers release the weights without the training data or code, and the license sets what you may do with the model. This library puts each model's full card, architecture, files, license and published evaluations on one page.
2,760Models
859Datasets
254Papers
1,692Publishers
5,040Sourced relationships
Updated 2026-09-18 · How the library is built
2,760 models, sorted by most downloaded.
VideoMAE model pre-trained on Kinetics-400 for 1600 epochs in a self-supervised way. It was introduced in the paper VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training by Tong et al. and first released in this repository. Disclaimer: The team releasing VideoMAE did not write a model card for this model so this model card has been written by the Hugging Face team. VideoMAE is an extension of Masked Autoencoders (MAE) to video. The architecture of the model is very similar to that of a standard Vision Transformer (ViT), with a decoder on top for predicting pixel values for masked patches. Videos are presented to the model as a sequence of…
This model can be used for the task of question answering. The model should not be used to intentionally create hostile or alienating environments for people. Significant research has explored bias and fairness issues with language models (see, e.g., Sheng et al. (2021) and Bender et al. (2021)). Predictions generated by the model may include disturbing and harmful stereotypes across protected classes; identity characteristics; and sensitive, social, and occupational groups. Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. The model creators note in the associated paper: The model creators note in the associated paper: The model…
Haystack is an AI orchestration framework to build customizable, production-ready LLM applications. You can use this model in Haystack to do extractive question answering on documents. To load and run the model with Haystack: For a complete example with an extractive question answering pipeline that scales over many documents, check out the corresponding Haystack tutorial. Evaluated on the SQuAD 2.0 English dev set with the official eval script. Evaluated on German MLQA: test-context-de-question-de.json Evaluated on German XQuAD: xquad.de.json For doing QA at scale (i.e. many docs instead of single paragraph), you can load the model also in haystack: Timo Möller: [email protected]…
This repository contains the Hugging Face Transformers conversion of the official VidEoMT checkpoint yt2019vitsmall52.8.pth from tue-mps/VidEoMT. The metrics above are the numbers reported by the authors in the official model zoo. Use processor.postprocessinstancesegmentation, processor.postprocesspanopticsegmentation, or processor.postprocesssemanticsegmentation depending on the target task.
Anime Face Detection using YOLOv8 Dataset was made created with 10 000 images manually annotated from safebooru. Split was train 70, val 20, test 10. This model is based on yolov8x6. It has been trained on the said dataset for 300 epoch at 1280px1280px. It took ~110 hours to train on a RTX A4000. On my dataset, the model performs particularly well with the default parameters. While it doesn't provide a huge mAP50-95, its predictions are always correct on the files I've tested. Confidence could be higher, but the model is very precise. Confusion matrix of the model, generated on the test set PR curve of the model, generated on the test set Other metrics are available in the./demo/val…
Mask2Former model trained on COCO panoptic segmentation (large-sized version, Swin backbone). It was introduced in the paper Masked-attention Mask Transformer for Universal Image Segmentation and first released in this repository. Disclaimer: The team releasing Mask2Former did not write a model card for this model so this model card has been written by the Hugging Face team. Mask2Former addresses instance, semantic and panoptic segmentation with the same paradigm: by predicting a set of masks and corresponding labels. Hence, all 3 tasks are treated as if they were instance segmentation. Mask2Former outperforms the previous SOTA, MaskFormer both in terms of performance an efficiency by (i)…
MaskFormer model trained on ADE20k semantic segmentation (base-sized version, Swin backbone). It was introduced in the paper Per-Pixel Classification is Not All You Need for Semantic Segmentation and first released in this repository. Disclaimer: The team releasing MaskFormer did not write a model card for this model so this model card has been written by the Hugging Face team. MaskFormer addresses instance, semantic and panoptic segmentation with the same paradigm: by predicting a set of masks and corresponding labels. Hence, all 3 tasks are treated as if they were instance segmentation. You can use this particular checkpoint for semantic segmentation. See the model hub to look for other…
ProcVLM-2B is a procedure-grounded vision-language model for estimating progress rewards from robot manipulation observations. Given a task description and a recent window of video frames, the model reasons about the remaining atomic actions and predicts the current task completion percentage. ProcVLM-2B is designed for research on robot learning, progress reward modeling, embodied evaluation, and procedure-aware video understanding. Typical use cases include: - estimating task completion progress from robot videos; - producing dense progress rewards from sparse demonstrations; - adapting progress prediction to a new environment with one-shot LoRA fine-tuning. This model is not intended to…
CLIPSeg model with reduce dimension 64. It was introduced in the paper Image Segmentation Using Text and Image Prompts by Lüddecke et al. and first released in this repository. This model is intended for zero-shot and one-shot image segmentation. Refer to the documentation.
NOTE Deprecated: use https://huggingface.co/tasksource/deberta-small-long-nli for longer context and better accuracy. This is DeBERTa-v3-base fine-tuned with multi-task learning on 600+ tasks of the tasksource collection. This checkpoint has strong zero-shot validation performance on many tasks (e.g. 70% on WNLI), and can be used for: - Zero-shot entailment-based classification for arbitrary labels [ZS]. - Natural language inference [NLI] - Hundreds of previous tasks with tasksource-adapters [TA]. - Further fine-tuning on a new task or tasksource task (classification, token classification or multiple-choice) [FT]. NLI training data of this model includes label-nli, a NLI dataset specially…
This repository contains the OpenVLA-OFT checkpoint for LIBERO-Spatial, as described in Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Success. OpenVLA-OFT significantly improves upon the base OpenVLA model by incorporating optimized fine-tuning techniques. See here for other OpenVLA-OFT checkpoints: https://huggingface.co/moojink?searchmodels=oft This example demonstrates generating an action chunk using a pretrained OpenVLA-OFT checkpoint. Ensure you have set up the conda environment as described in the GitHub README.
This is a ported version of The base model is wav2vec2-base, which is pretrained on 16kHz sampled speech audio. When using the model make sure that your speech input is also sampled at 16Khz. For more information refer to SUPERB: Speech processing Universal PERformance Benchmark Keyword Spotting (KS) detects preregistered keywords by classifying utterances into a predefined set of words. The task is usually performed on-device for the fast response time. Thus, accuracy, model size, and inference time are all crucial. SUPERB uses the widely used Speech Commands dataset v1.0 for the task. The dataset consists of ten classes of keywords, a class for silence, and an unknown class to include the…
Moirai, the Masked Encoder-based Universal Time Series Forecasting Transformer is a Large Time Series Model pre-trained on LOTSA data. For more details on the Moirai architecture, training, and results, please refer to the paper. Fig. 1: Overall architecture of Moirai. Visualized is a 3-variate time series, where variates 0 and 1 are target variables (i.e. to be forecasted, and variate 2 is a dynamic covariate (values in forecast horizon known). Based on a patch size of 64, each variate is patchified into 3 tokens. The patch embeddings along with sequence and variate id are fed into the Transformer. The shaded patches represent the forecast horizon to be forecasted, whose corresponding…
This checkpoint is a model fine-tuned for speech language identification (LID) and part of Facebook's Massive Multilingual Speech project. This checkpoint is based on the Wav2Vec2 architecture and classifies raw audio input to a probability distribution over 512 output classes (each class representing a language). The checkpoint consists of 1 billion parameters and has been fine-tuned from facebook/mms-1b on 512 languages. This MMS checkpoint can be used with Transformers to identify the spoken language of an audio. It can recognize the following 512 languages. Let's look at a simple example. First, we install transformers and some other libraries Note: In order to use MMS you need to have…
This repository contains the mT5 checkpoint finetuned on the 45 languages of XL-Sum dataset. For finetuning details and scripts, see the paper and the official repository. Scores on the XL-Sum test sets are as follows: Language | ROUGE-1 / ROUGE-2 / ROUGE-L Amharic | 20.0485 / 7.4111 / 18.0753 Arabic | 34.9107 / 14.7937 / 29.1623 Azerbaijani | 21.4227 / 9.5214 / 19.3331 Bengali | 29.5653 / 12.1095 / 25.1315 Burmese | 15.9626 / 5.1477 / 14.1819 Chinese (Simplified) | 39.4071 / 17.7913 / 33.406 Chinese (Traditional) | 37.1866 / 17.1432 / 31.6184 English | 37.601 / 15.1536 / 29.8817 French | 35.3398 / 16.1739 / 28.2041 Gujarati | 21.9619 / 7.7417 / 19.86 Hausa | 39.4375 / 17.6786 / 31.6667…
OneFormer model trained on the ADE20k dataset (large-sized version, Dinat backbone). It was introduced in the paper OneFormer: One Transformer to Rule Universal Image Segmentation by Jain et al. and first released in this repository. OneFormer is the first multi-task universal image segmentation framework. It needs to be trained only once with a single universal architecture, a single model, and on a single dataset, to outperform existing specialized models across semantic, instance, and panoptic segmentation tasks. OneFormer uses a task token to condition the model on the task in focus, making the architecture task-guided for training, and task-dynamic for inference, all with a single…
A 512×512 ONNX re-export of ZhengPeng7/BiRefNetlite that actually runs in a browser — solving the OOM wall that blocks every 1024×1024 variant from loading in onnxruntime-web. Drop it in with @huggingface/transformers to get high-quality alpha mattes entirely client-side, with no server round-trip. Used in production by Repper for per-motif matte refinement during foreground extraction. The 1024×1024 ONNX variants — including onnx-community/BiRefNetlite-ONNX — fail in every browser backend we tested: Root cause: BiRefNetlite's decoder produces very large intermediate tensors at 1024×1024 (multi-scale feature maps with 1024-way concatenations). The onnxruntime-web WASM heap is hardcoded at…
If you want to give the Finegrain Box Segmenter a try, the best way to is take a look at the Finegrain Object Cutter Space we shipped on Hugging Face: it's a fun "prompt to cut out" experience that will enable you to create pixel quality and high resolution cutouts for any object in a photo, by just naming the object. While building Finegrain, we needed a way to create pixel perfect and high resolution cutouts for objects in images. We looked at off-the-shelf solutions, but they simply didn't work for us: - On the one hand, traditional background removal models are great at producing HD cutouts, but unfortunately, different people will have different definitions for background and…
This checkpoint should be loaded into BartForConditionalGeneration.frompretrained. See the BART docs for more information.
Multi-Scale Efficient Global Context Vision Transformer (MS-EffGCViT) is a hybrid CNN-ViT architecture for deepfake detection. It fuses CNN-driven spatial inductive bias with hierarchical global-context attention to catch both local artifacts (textures, blending seams) and global artifacts (lighting, structural inconsistency). A single architecture ships in two sizes and three domain-tuned checkpoints, working on both static images and video at the frame level. - Frame-level — one model handles both images and videos (frame-level inference + aggregation). - Cross-domain — robust on both East-Asian (KoDF) and Western (Celeb-DF-v2, FaceForensics++) faces. - Two variants — Fast (b0) for…
Multi-Scale Efficient Global Context Vision Transformer (MS-EffGCViT) is a hybrid CNN-ViT architecture for deepfake detection. It fuses CNN-driven spatial inductive bias with hierarchical global-context attention to catch both local artifacts (textures, blending seams) and global artifacts (lighting, structural inconsistency). A single architecture ships in two sizes and three domain-tuned checkpoints, working on both static images and video at the frame level. - Frame-level — one model handles both images and videos (frame-level inference + aggregation). - Cross-domain — robust on both East-Asian (KoDF) and Western (Celeb-DF-v2, FaceForensics++) faces. - Two variants — Fast (b0) for…
This model has had its safety alignment substantially removed via abliteration (orthogonalizing the refusal direction out of the residual stream). It will comply with harmful, unethical, or illegal requests the original Qwen3.8-Flash-Next would refuse. Released strictly for legitimate research — interpretability, AI-safety / refusal-mechanism study, red-teaming, and robustness evaluation. You assume full responsibility for how you use it and everything it generates; add your own safety and moderation layers before any deployment. Use must comply with the Apache 2.0 License inherited from the base model and all applicable law. The authors accept no liability for misuse. - A Blackwell GPU…
Model Collections
Hand-picked starting points, each with the reason it exists.
Collection · 4 entries
Embedding models for retrieval
Sentence and document embedding models used to build retrieval systems. Dimension and sequence length matter more than size here, and both come from the publisher.
Collection · 4 entries
Models that fit on one accelerator
Models whose publisher-reported parameter count puts them within reach of a single accelerator at common precisions. Memory needed depends on precision and serving configuration, so treat the parameter count as the starting point, not the answer.
Collection · 6 entries
Open-weight text models worth knowing
Widely used open-weight language models, chosen because each one is a distinct family rather than a variant of the one above it. Selection, not a ranking.
Collection · 3 entries
Speech and audio models
Recognition and synthesis models, grouped so the two directions are easy to compare.
Open-Weight Models Explained
What is an open-weight model?
An AI model whose trained weights are published for anyone to download, so it can be run, tested and fine-tuned on hardware the user controls.
Is an open-weight model the same as open source?
Not always. Open weights means the trained model can be downloaded. Open source usually also means the training code and data are available and the license allows broad reuse. Many open-weight models release the weights only.
Can I use an open-weight model commercially?
It depends on the license. Apache 2.0 and MIT allow commercial use. Other licenses limit it, for example to non-commercial use or below a set number of users. Every model page here shows its license.
How much memory does an open-weight model need?
About two bytes per parameter at 16-bit precision, so a 7-billion-parameter model needs roughly 14 GB for its weights, plus memory for the context it processes. Each model page lists its parameter count and the size of its files.
Related SAVRN Research
The hub sits beside SAVRN's market data and infrastructure research: what models cost to run, and what it takes to run them.
SAVRN Index
What open models cost to run
The same open-weight model priced by every host that serves it, per million tokens.
Research Hub
Data center trackers and maps
Moratoriums, permits, power, water and capital behind the facilities that run these models.
Method
How the Model Hub is built
Sources, evidence labels, refresh behaviour, and the limits of every comparison here.