Open-Weight Models

Model · Image to 3d

TRELLIS.2-4B

Microsoft

TRELLIS.2 is a state-of-the-art large 3D generative model designed for high-fidelity image-to-3D generation. It leverages a novel "field-free" sparse voxel structure termed O-Voxel and a large-scale flow-matching transformer (4 Billion parameters). Unlike previous methods that rely on iso-surface fields (e.g., SDF, Flexicubes) which struggle with open surfaces or non-manifold geometry, TRELLIS can reconstruct and generate arbitrary 3D assets with complex topologies, sharp features, and full Physical-Based Rendering (PBR) materials—including transparency/translucency. - The CUDA Toolkit is needed to compile certain packages. Recommended version is 12.4. - Conda is recommended for managing…

This quants are specific for the DS4 inference engine. They may work with other inference engines or not (they should, but not the MTP model which requires a specific loader). https://github.com/antirez/ds4 Use q2 on 128 GB Mac machines, q4 on machines with ≥ 256 GB RAM, pair either with MTP for optional speculative decoding. The filename is the spec. In detail, for the q2 file: For the q4 file, only the three routed-expert classes change to Q4K. Everything else is byte-for-byte identical to the q2 recipe. The motivation behind the asymmetry: the routed experts are the majority of the parameter count but each individual expert handles only a fraction of tokens, so aggressive quantization on…

Model

docling-models

Docling

This page contains models that power the PDF document converion package docling. The layout model will take an image from a page and apply RT-DETR model in order to find different layout components. It currently detects the labels: Caption, Footnote, Formula, List-item, Page-footer, Page-header, Picture, Section-header, Table, Text, Title. As a reference (from the DocLayNet-paper), this is the performance of standard object detection methods on the DocLayNet dataset compared to human evaluation, The tableformer model will identify the structure of the table, starting from an image of a table. It uses the predicted table regions of the layout model to identify the tables. Tableformer has…

Model · Feature extraction

jina-embeddings-v3

Jina AI

jina-embeddings-v3 is a multilingual multi-task text embedding model designed for a variety of NLP applications. Based on the Jina-XLM-RoBERTa architecture, this model supports Rotary Position Embeddings to handle long input sequences up to 8192 tokens. Additionally, it features 5 LoRA adapters to generate task-specific embeddings efficiently. - retrieval.query: Used for query embeddings in asymmetric retrieval tasks - retrieval.passage: Used for passage embeddings in asymmetric retrieval tasks - separation: Used for embeddings in clustering and re-ranking applications - classification: Used for embeddings in classification tasks - text-matching: Used for embeddings in tasks that quantify…

Model · Text to audio

musicgen-medium

AI at Meta

MusicGen is a text-to-music model capable of genreating high-quality music samples conditioned on text descriptions or audio prompts. It is a single stage auto-regressive Transformer model trained over a 32kHz EnCodec tokenizer with 4 codebooks sampled at 50 Hz. Unlike existing methods, like MusicLM, MusicGen doesn't require a self-supervised semantic representation, and it generates all 4 codebooks in one pass. By introducing a small delay between the codebooks, we show we can predict them in parallel, thus having only 50 auto-regressive steps per second of audio. MusicGen was published in Simple and Controllable Music Generation by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant…

Model · Image classification

resnet50.a1_in1k

PyTorch Image Models

A ResNet-B image classification model. single layer 7x7 convolution with pooling 1x1 convolution shortcut downsample Trained on ImageNet-1k in timm using recipe template described below. ResNet Strikes Back A1 recipe - Deep Residual Learning for Image Recognition: https://arxiv.org/abs/1512.03385 Explore the dataset and runtime metrics of this model in timm model results.

Model · Image and text to text

Qwen3.5-35B-A3B

Qwen

Over recent months, we have intensified our focus on developing foundation models that deliver exceptional utility and performance. Qwen3.5 represents a significant leap forward, integrating breakthroughs in multimodal learning, architectural efficiency, reinforcement learning scale, and global accessibility to empower developers and enterprises with unprecedented capability and efficiency. For more details, please refer to our blog post Qwen3.5. WMT24++: a harder subset of WMT24 after difficulty labeling and rebalancing; we report the averaged scores on 55 languages using XCOMET-XXL. Empty cells (--) indicate scores not yet available or not applicable. Empty cells (--) indicate scores not…

Model · Speech recognition

romanian-wav2vec2

Théo Gigant

You can test this model online with the Space for Romanian Speech Recognition The model ranked TOP-1 on Romanian Speech Recognition during HuggingFace's Robust Speech Challenge: This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the Common Voice 8.0 - Romanian subset dataset, with extra training data from Romanian Speech Synthesis dataset. Without the 5-gram Language Model optimization, it achieves the following results on the evaluation set (Common Voice 8.0, Romanian subset, test split): The architecture is based on facebook/wav2vec2-xls-r-300m with a speech recognition CTC head and an added 5-gram language model (using pyctcdecode and kenlm) trained on the Romanian…

Model · Speech recognition

wav2vec2-large-voxrex-swedish

National Library of Sweden / KBLab

Finetuned version of KBs VoxRex large model using Swedish radio broadcasts, NST and Common Voice data. Evalutation without a language model gives the following: WER for NST + Common Voice test set (2% of total sentences) is 2.5%. WER for Common Voice test set is 8.49% directly and 7.37% with a 4-gram language model. When using this model, make sure that your speech input is sampled at 16kHz. This model has been fine-tuned for 120000 updates on NST + CommonVoice and then for an additional 20000 updates on CommonVoice only. The additional fine-tuning on CommonVoice hurts performance on the NST+CommonVoice test set somewhat and, unsurprisingly, improves it on the CommonVoice test set. It seems…

Model · Speech recognition

wav2vec2-large-xlsr-53-telugu

Anurag Singh

Fine-tuned facebook/wav2vec2-large-xlsr-53 on Telugu using the OpenSLR SLR66 dataset. When using this model, make sure that your speech input is sampled at 16kHz. The model can be used directly (without a language model) as follows: 70% of the OpenSLR Telugu dataset was used for training. Train Split of annotations is here Test Split of annotations is here Training Data Preparation notebook can be found here Training notebook can be foundhere Evaluation notebook is here

Model · Text generation

Qwen3.8-27B-Heretic-Abliterated-Uncensored-GGUF

Efe Buken

RVN is a double-refined abliterated variant of Qwen3.8-27B, built on top of (an ARA abliteration by Tim Rohrbaugh) and further refined with two additional full-weight ARA passes targeting residual refusals. It retains very low behavioral damage (KL ≈ 0.0085) while reducing harmful-prompt refusals from 3/100 (source) to 0–1/100 in independent measurements. ARA (Arbitrary-Rank Ablation) is the abliteration technique implemented in p-e-w/heretic. Traditional directional abliteration finds a single "refusal direction" in activation space and subtracts it — a one-shot, low-rank surgery that is simple but can leave residual refusals or damage unrelated behavior. ARA instead treats abliteration as…

Model · Image and text to text

Qwen3.5-27B

Qwen

Over recent months, we have intensified our focus on developing foundation models that deliver exceptional utility and performance. Qwen3.5 represents a significant leap forward, integrating breakthroughs in multimodal learning, architectural efficiency, reinforcement learning scale, and global accessibility to empower developers and enterprises with unprecedented capability and efficiency. For more details, please refer to our blog post Qwen3.5. WMT24++: a harder subset of WMT24 after difficulty labeling and rebalancing; we report the averaged scores on 55 languages using XCOMET-XXL. Empty cells (--) indicate scores not yet available or not applicable. Empty cells (--) indicate scores not…

Model · Text generation

GLM-4.7-Flash

Z.ai

Join our Discord community. Check out the GLM-4.7 technical blog, technical report(GLM-4.5). Use GLM-4.7-Flash API services on Z.ai API Platform. One click to GLM-4.7. GLM-4.7-Flash is a 30B-A3B MoE model. As the strongest model in the 30B class, GLM-4.7-Flash offers a new option for lightweight deployment that balances performance and efficiency. Default Settings (Most Tasks) For multi-turn agentic tasks (τ²-Bench and Terminal Bench 2), please turn on Preserved Thinking mode. Terminal Bench, SWE Bench Verified τ^2-Bench For τ^2-Bench evaluation, we added an additional prompt to the Retail and Telecom user interaction to avoid failure modes caused by users ending the interaction…

Model · Image and text to text

moondream2

Vik Korrapati

This repository contains the latest version of Moondream 2, our previous generation model. The latest version of Moondream is Moondream 3 (Preview). Moondream is a small vision language model designed to run efficiently everywhere. This repository contains the latest (2025-06-21) release of Moondream 2, as well as historical releases. The model is updated frequently, so we recommend specifying a revision as shown below if you're using it in a production application. Grounded Reasoning Introduces a new step-by-step reasoning mode that explicitly grounds reasoning in spatial positions within the image before answering, leading to more precise visual interpretation (e.g., chart median…

Model · Text generation

Llama-3.2-3B-Instruct

Meta Llama

The Llama 3.2 collection of multilingual large language models (LLMs) is a collection of pretrained and instruction-tuned generative models in 1B and 3B sizes (text in/text out). The Llama 3.2 instruction-tuned text only models are optimized for multilingual dialogue use cases, including agentic retrieval and summarization tasks. They outperform many of the available open source and closed chat models on common industry benchmarks. Model Architecture: Llama 3.2 is an auto-regressive language model that uses an optimized transformer architecture. The tuned versions use supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to align with human preferences for…

Model · Text to speech

chatterbox

Resemble AI

Chatterbox Multilingual V3 is the latest general-purpose multilingual TTS model in the Chatterbox family. It keeps the same 0.5B model size while improving speaker similarity, reducing hallucinations, and producing more natural, conversational speech across languages. V3 is designed for broad language coverage like V2, but with stronger stability and more expressive generation. It is the recommended multilingual model for users who want one voice cloning model that works across many languages. Try it in the Chatterbox Multilingual TTS V3 Space. Alongside V3, we are releasing the Single Language Pack: dedicated finetunes for priority languages where tighter quality control, stronger…

Model · Fill mask

bert-base-multilingual-cased

BERT community

Pretrained model on the top 104 languages with the largest Wikipedia using a masked language modeling (MLM) objective. It was introduced in this paper and first released in this repository. This model is case sensitive: it makes a difference between english and English. Disclaimer: The team releasing BERT did not write a model card for this model so this model card has been written by the Hugging Face team. BERT is a transformers model pretrained on a large corpus of multilingual data in a self-supervised fashion. This means it was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of publicly available data) with an automatic process to…

Model · Sentence similarity

nomic-embed-text-v2-moe

Nomic AI

This model was presented in the paper Training Sparse Mixture Of Experts Text Embedding Models. nomic-embed-text-v2-moe is a SoTA multilingual MoE text embedding model that excels at multilingual retrieval: Transformer-based text embedding models have improved their performance on benchmarks like MIRACL and BEIR by increasing their parameter counts. However, this scaling approach introduces significant deployment challenges, including increased inference latency and memory usage. These challenges are particularly severe in retrieval-augmented generation (RAG) applications, where large models' increased memory requirements constrain dataset ingestion capacity, and their higher latency…

Model · Time series forecasting

timesfm-2.5-200m-pytorch

Google

TimesFM (Time Series Foundation Model) is a pretrained time-series foundation model developed by Google Research for time-series forecasting. Please reinstall the latest version of the timesfm package to reflect these changes. Results should be unchanged. This checkpoint is not an officially supported Google product. See TimesFM in BigQuery for Google official support. timesfm-2.5-200m is the third open model checkpoint. timesfm-2.5-200m is pretrained using - Wikimedia Pageviews, cutoff Nov 2023 (see paper for details). - Google Trends top queries, cutoff EoY 2022 (see paper for details). - Synthetic and augmented data. At this point, please run

Model · Text generation

Gemma-4-26B-A4B-NVFP4

NVIDIA

Gemma 4 26B IT is an open multimodal model built by Google DeepMind that handles text and image inputs, can process video as sequences of frames, and generates text output. It is designed to deliver frontier-level performance for reasoning, agentic workflows, coding, and multimodal understanding on consumer GPUs and workstations, with a 256K-token context window and support for over 140 languages. The model uses a hybrid attention mechanism that interleaves local sliding-window and full global attention, with unified Keys and Values in global layers and Proportional RoPE (p-RoPE) to support long-context performance. The NVIDIA Gemma 4 26B IT NVFP4 model is quantized with NVIDIA Model…

Model · Speech recognition

nemotron-3.5-asr-streaming-0.6b-gguf

Handy

GGUF conversions of nvidia/nemotron-3.5-asr-streaming-0.6b for use with transcribe.cpp. Ported from upstream commit pinned 2026-06-08. Validated against the NeMo reference at transcribe.cpp commit Multilingual speech-to-text across 32 supported language-locales (the model's tokenizer recognizes 40, but 8 are adaptation-ready and need fine-tuning) with punctuation and capitalization. A 0.6B-parameter cache-aware streaming FastConformer encoder with a prompt-conditioned RNN-T transducer decoder; the target language is selected per call (--language en-US, fr-FR, de-DE,...) and an auto mode emits a tag. Ships both the offline path (attcontextsize=[56, 13], 1.12s, headline accuracy) and…

Model · Text generation

Qwen2.5-Coder-14B-Instruct

Qwen

Qwen2.5-Coder is the latest series of Code-Specific Qwen large language models (formerly known as CodeQwen). As of now, Qwen2.5-Coder has covered six mainstream model sizes, 0.5, 1.5, 3, 7, 14, 32 billion parameters, to meet the needs of different developers. Qwen2.5-Coder brings the following improvements upon CodeQwen1.5: - Significantly improvements in code generation, code reasoning and code fixing. Base on the strong Qwen2.5, we scale up the training tokens into 5.5 trillion including source code, text-code grounding, Synthetic data, etc. Qwen2.5-Coder-32B has become the current state-of-the-art open-source codeLLM, with its coding abilities matching those of GPT-4o. - A more…

Model · Text ranking

ms-marco-MiniLM-L12-v2

Sentence Transformers - Cross-Encoders

This model was trained on the MS Marco Passage Ranking task. The model can be used for Information Retrieval: Given a query, encode the query will all possible passages (e.g. retrieved with ElasticSearch). Then sort the passages in a decreasing order. See SBERT.net Retrieve & Re-rank for more details. The training code is available here: SBERT.net Training MS Marco The usage is easy when you have SentenceTransformers installed. Then you can use the pre-trained models like this: In the following table, we provide various pre-trained Cross-Encoders together with their performance on the TREC Deep Learning 2019 and the MS Marco Passage Reranking dataset.

Model · Image and text to text

gemma-4-26B-A4B-it-AWQ-4bit

Cyankiwi

Gemma is a family of open models built by Google DeepMind. Gemma 4 models are multimodal, handling text and image input (with audio supported on small models) and generating text output. This release includes open-weights models in both pre-trained and instruction-tuned variants. Gemma 4 features a context window of up to 256K tokens and maintains multilingual support in over 140 languages. Featuring both Dense and Mixture-of-Experts (MoE) architectures, Gemma 4 is well-suited for tasks like text generation, coding, and reasoning. The models are available in four distinct sizes: E2B, E4B, 26B A4B, and 31B. Their diverse sizes make them deployable in environments ranging from high-end phones…