For more information, check out the official repository and example colab. If you haven't already, you can install the Transformers.js JavaScript library from NPM using: You can then use the model for portrait matting, as follows: Or with the AutoModel and AutoProcessor APIs: Note: Having a separate repo for ONNX weights is intended to be a temporary solution until WebML gains more traction. If you would like to make your models web-ready, we recommend converting to ONNX using Optimum and structuring your repo like this one (with ONNX weights located in a subfolder named onnx).
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.
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 Emotion Recognition (ER) predicts an emotion class for each utterance. The most widely used ER dataset IEMOCAP is adopted, and we follow the conventional evaluation protocol: we drop the unbalanced emotion classes to leave the final four classes with a similar amount of data points and cross-validate on five folds of the standard splits. For the original model's training and evaluation instructions refer to the You…
basemodel: stabilityai/stable-diffusion-xl-base-1.0 - stable-diffusion-xl - stable-diffusion-xl-diffusers - text-to-image - diffusers - inpainting SD-XL Inpainting 0.1 is a latent text-to-image diffusion model capable of generating photo-realistic images given any text input, with the extra capability of inpainting the pictures by using a mask. The SD-XL Inpainting 0.1 was initialized with the stable-diffusion-xl-base-1.0 weights. The model is trained for 40k steps at resolution 1024x1024 and 5% dropping of the text-conditioning to improve classifier-free classifier-free guidance sampling. For inpainting, the UNet has 5 additional input channels (4 for the encoded masked-image and 1 for the…
This is the FP8 mixed-precision quantization of HiDream-O1-Image for use with ComfyUI. By quantizing to 8-bit floats, the model fits comfortably within ~10 GB of VRAM — making it accessible on 12 GB GPUs (RTX 3080/4070/4080, etc.) with minimal quality trade-off. This is the recommended variant for GPUs with less than 16 GB VRAM. Tested on 12 GB cards at 2048 × 2048 resolution. Or install via ComfyUI Manager by searching for HiDream O1. Open ComfyUI and use the workflow provided in the custom node repository. Point the model loader to HiDream-O1-Image-fp8. HiDream-O1-Image is a natively unified image generative foundation model built on a Pixel-level Unified Transformer (UiT) — no external…
Higgs TTS 3 is built for voice chat: it speaks, not just reads. It turns model responses into expressive conversational speech across 100+ languages, with zero-shot voice cloning and inline control over emotion, style, prosody, pauses, and sound effects. Higgs autoregressive decoder consumes interleaved text and audio tokens. Audio is encoded by the Higgs Tokenizer into 8 codebooks at 25 fps, staggered via a delay pattern, then mapped to backbone hidden states through a multi-codebook fused embedding. Output codes pass through a multi-codebook fused head, are de-delayed, and decoded back to waveform. The model reaches single-digit WER/CER on 102 languages, which split into · Chichewa/Nyanja…
A Vision Transformer (ViT) image classification model. Trained on ImageNet-21k and fine-tuned on ImageNet-1k (with additional augmentation and regularization) in JAX by paper authors, ported to PyTorch by Ross Wightman. - How to train your ViT? Data, Augmentation, and Regularization in Vision Transformers: https://arxiv.org/abs/2106.10270 - An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale: https://arxiv.org/abs/2010.11929v2 Explore the dataset and runtime metrics of this model in timm model results.
OneFormer model trained on the Cityscapes dataset (large-sized version, Swin 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 ConvNeXt-V2 image classification model. Pretrained with a fully convolutional masked autoencoder framework (FCMAE) and fine-tuned on ImageNet-1k. Explore the dataset and runtime metrics of this model in timm model results. All timing numbers from eager model PyTorch 1.13 on RTX 3090 w/ AMP.
+ Always use full LoRa name with "vrtlxxxx" trigger in prompt like "Alba Baptista (vrtlalbabaptista) in a swimming pool". "Woman" or "1girl" will NOT work due to my way i do captions. + Add the gender to the prompt for confusing names like "Alex Jones". + Remove the name when internal model knowledge is bad or censored or is confusing to model like "Sandy Cheeks" or "Kate Middleton". + When using a Lora with multiple triggers (vrtlxx,vrtlyy) do not use the real character name but only trigger or a combination of it. "vrtlMain" always combines all trigger-words. Angourie Rice, January Jones, Julianna Guill, Ursula Corbero, Judith Rakers, Alina Merkau, Kiernan Shipka, Leslie Bibb, Marie…
This is mbart-large-cc25, finetuned on wmtenro. It scores BLEU 28.1 without post processing and BLEU 38 with postprocessing. Instructions in romanianpostprocessing.md Original Code: https://github.com/pytorch/fairseq/tree/master/examples/mbart Docs: https://huggingface.co/transformers/master/modeldoc/mbart.html
This model takes xlm-roberta-large and fine-tunes it on a combination of NLI data in 15 languages. It is intended to be used for zero-shot text classification, such as with the Hugging Face ZeroShotClassificationPipeline. This model is intended to be used for zero-shot text classification, especially in languages other than English. It is fine-tuned on XNLI, which is a multilingual NLI dataset. The model can therefore be used with any of the languages in the XNLI corpus: Since the base model was pre-trained trained on 100 different languages, the model has shown some effectiveness in languages beyond those listed above as well. See the full list of pre-trained languages in appendix A of the…
A ResNet-D image classification model. 3-layer stem of 3x3 convolutions with pooling 2x2 average pool + 1x1 convolution shortcut downsample Trained on ImageNet-1k in timm using recipe template described below. RandAugment RA2 recipe. Inspired by and evolved from EfficientNet RandAugment recipes. Published as B recipe in ResNet Strikes Back. RMSProp (TF 1.0 behaviour) optimizer, EMA weight averaging Step (exponential decay w/ staircase) LR schedule with warmup - Deep Residual Learning for Image Recognition: https://arxiv.org/abs/1512.03385 - Bag of Tricks for Image Classification with Convolutional Neural Networks: https://arxiv.org/abs/1812.01187 Explore the dataset and runtime metrics of…
Mask2Former model trained on COCO instance segmentation (tiny-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)…
a sentence initial language token is required in the form of >>id<< (id = valid target language ID) - hfname: eng-zho - sourcelanguages: eng - targetlanguages: zho - opusreadmeurl: https://github.com/Helsinki-NLP/Tatoeba-Challenge/tree/master/models/eng-zho/README.md - originalrepo: Tatoeba-Challenge - srcconstituents: {'eng'} - tgtconstituents: {'cmnHans', 'nan', 'nanHani', 'gan', 'yue', 'cmnKana', 'yueHani', 'wuuBopo', 'cmnLatn', 'yueHira', 'cmnHani', 'cjyHans', 'cmn', 'lzhHang', 'lzhHira', 'cmnHant', 'lzhBopo', 'zho', 'zhoHans', 'zhoHant', 'lzhHani', 'yueHang', 'wuu', 'yueKana', 'wuuLatn', 'yueBopo', 'cjyHant', 'yueHans', 'lzh', 'cmnHira', 'lzhYiii', 'lzhHans', 'cmnBopo', 'cmnHang'…
This model is a conversion of MoritzLaurer/roberta-base-zeroshot-v2.0-c to ONNX format using the Optimum library.
FLUX.2 [klein] 4B Base is a 4 billion parameter rectified flow transformer capable of generating images from text descriptions and supports multi-reference editing capabilities. For more information, please read our blog post. This repository holds an FP8 version of FLUX.2 [klein] 4B Base. The main repository of this model (full BF16 weights) can be found here. Limitations - This model is not intended or able to provide factual information. - While the model can output text, text rendered may be inaccurate or subject to distortion. - As a statistical model, this checkpoint may represent or amplify biases observed in the training data. - The model may fail to generate output that matches the…
X-CLIP model (base-sized, patch resolution of 16) trained on Kinetics-400. It was introduced in the paper Expanding Language-Image Pretrained Models for General Video Recognition by Ni et al. and first released in this repository. This model was trained using 32 frames per video, at a resolution of 224x224. Disclaimer: The team releasing X-CLIP did not write a model card for this model so this model card has been written by the Hugging Face team. X-CLIP is a minimal extension of CLIP for general video-language understanding. The model is trained in a contrastive way on (video, text) pairs. This allows the model to be used for tasks like zero-shot, few-shot or fully supervised video…
A french sequence to sequence pretrained model based on BART. BARThez is pretrained by learning to reconstruct a corrupted input sentence. A corpus of 66GB of french raw text is used to carry out the pretraining. Unlike already existing BERT-based French language models such as CamemBERT and FlauBERT, BARThez is particularly well-suited for generative tasks (such as abstractive summarization), since not only its encoder but also its decoder is pretrained. In addition to BARThez that is pretrained from scratch, we continue the pretraining of a multilingual BART mBART which boosted its performance in both discriminative and generative tasks. We call the french adapted version mBARThez.
This is an NVFP4 quantized version of Qwen3-VL-8B-Instruct, a powerful vision-language model for multimodal understanding and generation tasks. The following modules were excluded from quantization to maintain model quality: - lmhead (language model head) - Visual encoder modules (model.visual.) - MLP gate projections (.mlp.gate$) For faster inference, you can use this model with vLLM: This quantized model maintains high quality for vision-language tasks while significantly reducing memory usage. The SmoothQuant technique helps preserve model accuracy during quantization. Typical quality degradation is 2-5% compared to the full-precision model. 1. Calibration: Used 512 samples from the…
This model is a core component of the meikiocr pipeline. For the full implementation, command-line script, and documentation, please see the official GitHub repository. pareto-optimal text recognition model. trained on japanese video games. meiki.text.recognition achieves state-of-the-art text recognition accuracy as well as latency by redefining "text recognition" as "character detection". the model is a fine-tune of https://github.com/Peterande/D-FINE object detecor combined with a mobilenetv4 CNN backbone. to my knowledge there is no existing, open weight text recognition model with a better accuracy/latency tradeoff for japanese text recognition. - it is specifically trained on japanese…
lykon/dreamshaper-8 is a Stable Diffusion model that has been fine-tuned on runwayml/stable-diffusion-v1-5. For more general information on how to run text-to-image models with Diffusers, see the docs. - Version 8 focuses on improving what V7 started. Might be harder to do photorealism compared to realism focused models, as it might be hard to do anime compared to anime focused models, but it can do both pretty well if you're skilled enough. Check the examples! - Version 7 improves lora support, NSFW and realism. If you're interested in "absolute" realism, try AbsoluteReality. - Version 6 adds more lora support and more style in general. It should also be better at generating directly at…
A ConvNeXt image classification model. Pretrained in timm on ImageNet-12k (a 11821 class subset of full ImageNet-22k) and fine-tuned on ImageNet-1k by Ross Wightman. ImageNet-12k training done on TPUs thanks to support of the TRC program. Fine-tuning performed on 8x GPU Lambda Labs cloud instances. Explore the dataset and runtime metrics of this model in timm model results. All timing numbers from eager model PyTorch 1.13 on RTX 3090 w/ AMP.
Run with https://llama.app - https://huggingface.co/google/gemma-4-E2B-it - https://huggingface.co/google/gemma-4-E2B-it-assistant - https://huggingface.co/google/gemma-4-E2B-it-qat-q40-unquantized-assistant - https://huggingface.co/google/gemma-4-E2B-it-qat-q40-unquantized - add info - add dflash
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.