전체 검색

AI 모델 2119

• wav2vec2-large-xlsr-lithuanian↗

  m3hrdadfi

  Fine-tuned facebook/wav2vec2-large-xlsr-53 in Lithuanian using Common Voice. When using this model, make sure that your speech input is sampled at 16kHz. You can see the training states here The script used for training can be found here

• gemma-3-12b-it-abliterated-v2-GGUF↗

  mlabonne

  This is an uncensored version of google/gemma-3-12b-it created with a new abliteration technique. See this article to know more about abliteration. The refusal direction is computed by comparing the residual streams between target (harmful) and baseline (harmless) samples. The hidden states of target modules (e.g., o_proj) are orthogonalized to subtract this refusal direction with a given weight factor. These weight factors follow a normal distribution with a certain spread and peak layer. Modules can be iteratively orthogonalized in batches, or the refusal direction can be accumulated to save memory. Finally, I used a hybrid evaluation with a dedicated test set to calculate the acceptance rate. This uses both a dictionary approach and NousResearch/Minos-v1. The goal is to obtain an acceptance rate >90% and still produce coherent outputs.

• resnet34.a1_in1k↗

  timm

  This model features: * ReLU activations * single layer 7x7 convolution with pooling * 1x1 convolution shortcut downsample Recipe details: * ResNet Strikes Back `A1` recipe * LAMB optimizer with BCE loss * Cosine LR schedule with warmup - **Model Type:** Image classification / feature backbone - **Model Stats:** - Params (M): 21.8 - GMACs: 3.7 - Activations (M): 3.7 - Image size: train = 224 x 224, test = 288 x 288 - **Papers:** - ResNet strikes back: An improved training procedure in timm: https://arxiv.org/abs/2110.00476 - Deep Residual Learning for Image Recognition: https://arxiv.org/abs/1512.03385 - **Original:** https://github.com/huggingface/pytorch-image-models

• voxcpm2-GGUF↗

  cstr

  Historical T4 Vulkan component measurements found CFM per audio step of 70.2 ms for Q8_0 versus 59.4 ms for the mixed recipe at ten steps. An eight-step run measured 46.7 ms. These component timings do **not** establish speech quality or an accepted end-to-end speedup. The mixed file was built with `crispasr-quantize voxcpm2-f16.gguf out.gguf q8_0 --tensor-type "^locdit\.=f16"`. A previous CPU synthesis roundtrip is not GPU acceptance. **Validation update, 2026-10-09:** matched real T4 tests at ten steps and seed 2 produced repeated tokens in the **Nemotron ASR transcripts** of the short test sentence for both Q8_0 and the mixed recipe. Six of sixteen exact primary ASR cases passed. This rejected gate is not proof that synthesized audio audibly stutters: a later fixed-audio comparison found **both Parakeet and Qwen3 transcribe every one of 22 original Python/native CPU/GPU controls exactly** across three resampling paths, while Nemotron retains errors. Resampling alone does not resolve the discrepancy. The completed original NVIDIA float32 and native F16 comparison reads **all 22 VoxCPM2 controls correctly**; native Q4 reads only 8/22. Original and F16 normalized words agree on all 26 tested clips, and fresh/reused sessions agree. The repeated tokens are therefore attributable to the native Nemotron Q4 recognizer on these fixed controls. Critical RNNT precision guards are being tested before changing that quant or its defaults. The published Q4 primary gate remains rejected while that recognizer is repaired. This diagnosis does not certify eight steps, Intel B390 performance, or complete original/native waveform parity. Defaults remain ten steps. Original NVIDIA/F16/Q4 inputs, features, encoder states and complete transcripts (SHA256 `7762f2a73d8ae7dad05d667fbf0bca393a0b3b7ad622b1b1f09473fa8f5d0463`).

• sew-d-tiny-100k-ft-ls100h↗

  optimum-intel-internal-testing

  The base model pretrained on 16kHz sampled speech audio. When using the model make sure that your speech input is also sampled at 16Khz. Note that this model should be fine-tuned on a downstream task, like Automatic Speech Recognition, Speaker Identification, Intent Classification, Emotion Recognition, etc... **Abstract** This paper is a study of performance-efficiency trade-offs in pre-trained models for automatic speech recognition (ASR). We focus on wav2vec 2.0, and formalize several architecture designs that influence both the model performance and its efficiency. Putting together all our observations, we introduce SEW (Squeezed and Efficient Wav2vec), a pre-trained model architecture with significant improvements along both performance and efficiency dimensions across a variety of training setups. For example, under the 100h-960h semi-supervised setup on LibriSpeech, SEW achieves a 1.9x inference speedup compared to wav2vec 2.0, with a 13.5% relative reduction in word error rate. With a similar inference time, SEW reduces word error rate by 25-50% across different model sizes. This code snippet shows how to evaluate **asapp/sew-d-tiny-100k-ft-ls100h** on LibriSpeech's "clean" and "other" test data.

• Depth-Anything-V2-Small-hf↗

  depth-anything

  Depth Anything V2 is trained from 595K synthetic labeled images and 62M+ real unlabeled images, providing the most capable monocular depth estimation (MDE) model with the following features: - more fine-grained details than Depth Anything V1 - more robust than Depth Anything V1 and SD-based models (e.g., Marigold, Geowizard) - more efficient (10x faster) and more lightweight than SD-based models - impressive fine-tuned performance with our pre-trained models Depth Anything V2 was introduced in the paper of the same name by Lihe Yang et al. It uses the same architecture as the original Depth Anything release, but uses synthetic data and a larger capacity teacher model to achieve much finer and robust depth predictions. The original Depth Anything model was introduced in the paper Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data by Lihe Yang et al., and was first released in this repository. Depth Anything V2 leverages the DPT architecture with a DINOv2 backbone.

• maskformer-swin-base-ade↗

  facebook

  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.

• Samantha-1.1-70b-i1-GGUF↗

  mradermacher

• BiomedVLP-BioViL-T↗

  Microsoft

  BioViL-T is a domain-specific vision-language model designed to analyze chest X-rays (CXRs) and radiology reports. It was trained using a temporal multi-modal pre-training procedure, which distinguishes it from its predecessor model (BioViL). In detail, BioViL-T takes advantage of the temporal structure between data points, resulting in improved downstream performance on multiple benchmarks, while using the same training dataset as its predecessor. In particular, the resultant model displays significant improvement in embedding temporal information present in the image and text modalities (see results), as well as in the joint space. The canonical model can be adapted to both single- and multi-image downstream applications including: natural language inference, phrase-grounding, image/text classification, and language decoding. The corresponding BERT language model is trained in two stages: First, we pretrain CXR-BERT-general from a randomly initialized BERT model via Masked Language Modeling (MLM) on PubMed abstracts and clinical notes from the publicly-available MIMIC-III and MIMIC-CXR. The general model can be fine-tuned for research in other clinical domains by adjusting the parameters specific to the target domain. In the second stage, BioViL-T is continually pretrained from CXR-BERT-general using a multi-modal pre-training procedure by utilising radiology reports and sequences of chest X-rays. We utilise the latent representation of [CLS] token to align text and image embeddings. The image model is jointly trained with the text model in a multi-modal contrastive learning framework. It's a hybrid image encoder composed of a Vision Transformer and ResNet-50, where the latter is used as backbone network to extract features from images at each time point. The transformer is included in the design to aggregate and compare image features extracted across the temporal dimension. The corresponding model definition and its loading functions can be accessed through our HI-ML-Multimodal GitHub repository. The joint image and text model, namely BioViL-T, can be used in phrase grounding applications as shown in this python notebook example. Additionally, please check t

• Nemotron-3.5-Embed-1B-BF16↗

  NVIDIA

  **Nemotron-3.5-Embed-1B-BF16** is a versatile embedding model developed by NVIDIA for multimodal and multilingual question-answering retrieval. The model can embed a document page in the form of an image, text, or combined image–text input. Documents can be retrieved given a user query in text form. The model supports page images containing text, tables, charts, and infographics. For text retrieval, the model was evaluated across 34 languages: English, Arabic, Assamese, Bengali, Bulgarian, Chinese, Danish, Dutch, Finnish, French, German, Hindi, Hinglish, Indonesian, Italian, Japanese, Korean, Malay, Marathi, Nepali, Norwegian, Persian, Portuguese, Romanian, Russian, Spanish, Swahili, Swedish, Tamil, Telugu, Thai, Ukrainian, Urdu, and Vietnamese. For image retrieval, it was evaluated across 18 languages: Arabic, Bengali, Chinese, English, Persian, Finnish, French, German, Hindi, Indonesian, Japanese, Korean, Russian, Spanish, Swahili, Telugu, Thai, and Yoruba. The model may support additional languages beyond these evaluation sets. Among models of comparable size, Nemotron-3.5-Embed-1B-BF16 achieves state-of-the-art results across multiple multilingual and multimodal retrieval benchmarks. Read more details in our Blog Post. The **Nemotron-3.5-Embed-1B-BF16** is suitable for users who want to build a multimodal question-and-answer application over a large corpus, leveraging the latest dense retrieval technology. The input of the model is a text or a document image, and the output is a fixed-size embedding vector. The embedding model is a bi-encoder that supports context in textual format (e.g. the query or the OCR text of a page or a section of a document) or the image of a document page. Typically, the embedding model is used first to embed (vectorize) the whole corpus (document images or text chunks), and embeddings are stored in a vector database associated to its raw content (image or text). Then at inference time, the embedding model is used to embed the query. The model generates dense vector embeddings from text or image inputs, enabling multimodal retrieval, semantic search, and (agentic) RAG workflows.