サイト内検索

AI モデル 2119

• gemma-2-2b-it↗

  Efficient-Large-Model

  * [Responsible Generative AI Toolkit][rai-toolkit] * [Gemma on Kaggle][kaggle-gemma] * [Gemma on Vertex Model Garden][vertex-mg-gemma2] Gemma is a family of lightweight, state-of-the-art open models from Google, built from the same research and technology used to create the Gemini models. They are text-to-text, decoder-only large language models, available in English, with open weights for both pre-trained variants and instruction-tuned variants. Gemma models are well-suited for a variety of text generation tasks, including question answering, summarization, and reasoning. Their relatively small size makes it possible to deploy them in environments with limited resources such as a laptop, desktop or your own cloud infrastructure, democratizing access to state of the art AI models and helping foster innovation for everyone. Below we share some code snippets on how to get quickly started with running the model. First, install the Transformers library with:

• ThinkPRM-1.5B↗

  launch

  ThinkPRM-1.5B is a generative Process Reward Model (PRM) based on the R1-Distill-Qwen-1.5B architecture. It is fine-tuned to perform step-by-step verification of reasoning processes (like mathematical solutions) by generating an explicit verification chain-of-thought (CoT) that involves labeling every step. It is designed to be highly data-efficient, requiring significantly less supervision data than traditional discriminative PRMs while achieving strong performance. ThinkPRM-1.5B provides step-level verification scores by generating natural language critiques and correctness judgments for each step in a given solution prefix. It leverages the underlying reasoning capabilities of the base Large Reasoning Model (LRM) and enhances them through fine-tuning on a small (1K examples) dataset of synthetically generated verification CoTs. These synthetic CoTs were produced by prompting QwQ-32B-Preview and filtered against ground-truth step labels from the PRM800K dataset to ensure quality. The model uses a standard language modeling objective, making it interpretable and allowing it to scale process verification compute by generating longer or multiple verification CoTs. It demonstrated superior performance compared to LLM-as-a-judge and discriminative PRM baselines (based on the same R1-Distill-Qwen-1.5B model but trained on ~100x more labels) on benchmarks including ProcessBench, MATH-500, AIME '24, GPQA-Diamond, and LiveCodeBench.

• deformable-detr-DocLayNet↗

  Aryn

  Deformable DEtection TRansformer (DETR), trained on DocLayNet (including 80k annotated pages in 11 classes). The DETR model is an encoder-decoder transformer with a convolutional backbone. Two heads are added on top of the decoder outputs in order to perform object detection: a linear layer for the class labels and a MLP (multi-layer perceptron) for the bounding boxes. The model uses so-called object queries to detect objects in an image. Each object query looks for a particular object in the image. For COCO, the number of object queries is set to 100. The model is trained using a "bipartite matching loss": one compares the predicted classes + bounding boxes of each of the N = 100 object queries to the ground truth annotations, padded up to the same length N (so if an image only contains 4 objects, 96 annotations will just have a "no object" as class and "no bounding box" as bounding box). The Hungarian matching algorithm is used to create an optimal one-to-one mapping between each of the N queries and each of the N annotations. Next, standard cross-entropy (for the classes) and a linear combination of the L1 and generalized IoU loss (for the bounding boxes) are used to optimize the parameters of the model.

• zero123-xl-diffusers↗

  ashawkey

  _Note: This section is originally taken from the Stable Diffusion v2 model card, but applies in the same way to Zero-1-to-3._ The model is intended for research purposes only. Possible research areas and tasks include: - Safe deployment of large-scale models. - Probing and understanding the limitations and biases of generative models. - Generation of artworks and use in design and other artistic processes. - Applications in educational or creative tools. - Research on generative models.

• umt5-base↗

  Google

  UMT5 is pretrained on the an updated version of mC4 corpus, covering 107 languages: Afrikaans, Albanian, Amharic, Arabic, Armenian, Azerbaijani, Basque, Belarusian, Bengali, Bulgarian, Burmese, Catalan, Cebuano, Chichewa, Chinese, Corsican, Czech, Danish, Dutch, English, Esperanto, Estonian, Filipino, Finnish, French, Galician, Georgian, German, Greek, Gujarati, Haitian Creole, Hausa, Hawaiian, Hebrew, Hindi, Hmong, Hungarian, Icelandic, Igbo, Indonesian, Irish, Italian, Japanese, Javanese, Kannada, Kazakh, Khmer, Korean, Kurdish, Kyrgyz, Lao, Latin, Latvian, Lithuanian, Luxembourgish, Macedonian, Malagasy, Malay, Malayalam, Maltese, Maori, Marathi, Mongolian, Nepali, Norwegian, Pashto, Persian, Polish, Portuguese, Punjabi, Romanian, Russian, Samoan, Scottish Gaelic, Serbian, Shona, Sindhi, Sinhala, Slovak, Slovenian, Somali, Sotho, Spanish, Sundanese, Swahili, Swedish, Tajik, Tamil, Telugu, Thai, Turkish, Ukrainian, Urdu, Uzbek, Vietnamese, Welsh, West Frisian, Xhosa, Yiddish, Yoruba, Zulu. **Note**: UMT5 was only pre-trained on mC4 excluding any supervised training. Therefore, this model has to be fine-tuned before it is useable on a downstream task.

• Mage-VL↗

  Microsoft

  **Mage-VL** is a **codec-native, proactive-streaming multimodal foundation model** for image and video understanding, whose visual encoder is trained **entirely from scratch** at a compact **4B** scale. It targets a modern *Moravec's paradox* of VLMs — strong at complex offline reasoning, yet slow and compute-heavy on simple real-time streaming perception. Instead of decoding video into uniformly-sampled frames and pushing a dense grid of patch tokens through a frozen web-pretrained ViT, Mage-VL follows the structure of modern video codecs: it separates a stream into **anchor (I) frames** and **predicted (P) frames**, keeps every anchor patch, and retains only the predicted-frame patches where the codec spends bits — the regions carrying real motion and new detail. This codec-aligned sparsity cuts visual tokens by **over 75%** while preserving spatio-temporal context, yielding **up to 3.5× wall-clock inference speedup** over uniform frame sampling. On top of this pair, a **System 1 & System 2 dual-process design** adds proactive streaming inside a single model: a lightweight **cognition gate** (System 1) watches each rolling codec window and stays silent on routine content, invoking the full VLM (System 2) only when a response-worthy event completes — no multi-agent pipeline required. A **single checkpoint**, `microsoft/Mage-VL`, is one unified model that **simultaneously** provides image & video understanding **and** the proactive streaming gate — the same weights answer offline image/video questions and drive event-gated commentary. It covers every Mage-VL capability: image understanding, frame-sampled video, traditional H.264/HEVC codec video, neural DCVC-RT codec video, and event-gated streaming. The repository bundles the codec processor, the neural codec package, and the proactive gate weights — no separate understanding, NVC, or streaming checkpoint is required.

• robeczech-base↗

  ufal

  The tokenizer in the initial release (a) contained a hole (51959 did not correspond to any token), and (b) mapped several tokens (unseen during training but required by the BBPE tokenizer) to the same ID as the `[UNK]` token (3). That sometimes caused problems, as in https://huggingface.co/ufal/robeczech-base/discussions/4. See https://huggingface.co/ufal/robeczech-base/discussions/4#64b8f6a7f1f8e6ea5860b314 for more information. In version 1.1, the tokenizer was modified by (a) removing the hole, (b) mapping all tokens to a unique ID. That also required increasing the vocabulary size and embeddings weights (by replicating the embedding of the `[UNK]` token). Without finetuning, version 1.1 and version 1.0 gives exactly the same embeddings on any input (apart from the pooler missing in v1.1), and the tokens in version 1.0 that mapped to a different ID than the `[UNK]` token map to the same ID in version 1.1. However, the sizes of the embeddings (and LM head weights and biases) are different, so the weights of the version 1.1 are not compatible with the configuration of version 1.0 and vice versa.

• gemma-4-12B-it-AWQ-INT4↗

  cyankiwi

  Hugging Face | GitHub | Launch Blog | Documentation | Technical Report > [!Note] > This model card is for the Gemma 4 12B Unified model, which is part of the Gemma 4 family of open models. Built with the same multimodal functionality as Gemma 4 E2B and E4B (text, audio, image, and video inputs), it brings native audio and vision understanding directly to local environments without the need for separate encoders. This unified approach to multimodality makes the model encoder-free, offering a deployment size that is perfect for consumer devices and streamlined local execution. 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 E2B, E4B, and 12B) 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.

• gemma-4-31B-it-qat-q4_0-gguf↗

  Google

  Hugging Face | GitHub | Launch Blog | Documentation | Technical Report > [!Note] > This model card is for the new versions of the Gemma 4 family optimized with Quantization-Aware Training (QAT), which allows preserving similar quality to bfloat16 while dramatically reducing the memory requirements to load the model. > Four versions of the QAT checkpoints are available: > * **Unquantized QAT checkpoints** (Q4_0): Half-precision weights extracted from the QAT pipeline, ideal for custom downstream compilation and research. Available for Gemma 4 E2B, E4B, 12B, 26B A4B, and 31B, and their drafter models. > * **GGUF** (Q4_0): Ready-to-deploy formats for broad ecosystem compatibility. Available for Gemma 4 E2B, E4B, 12B, 26B A4B, and 31B. > * **Mobile-optimized** (wNa8o8): A custom schema engineered explicitly for mobile hardware efficiency. It features targeted 2-bit decoding layers, optimized KV caches, and static activations to maximize VRAM savings. Available for Gemma 4 E2B and E4B. > * **Compressed Tensors** (w4a16): QAT checkpoints serialized in the compressed-tensors format for native, optimized inference with vLLM. Available for Gemma 4 E2B, E4B, 12B, and 31B. > * **Assistant Compatibility**: When using multi-token prediction (speculative decoding) with an assistant model alongside a QAT target model, the assistant model must also be a QAT checkpoint with the same precision to ensure compatibility. 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 E2B, E4B, and 12B) 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.

• gemma-3-1b-it-unsloth-bnb-4bit↗

  unsloth

  Gemma is a family of lightweight, state-of-the-art open models from Google, built from the same research and technology used to create the Gemini models. Gemma 3 models are multimodal, handling text and image input and generating text output, with open weights for both pre-trained variants and instruction-tuned variants. Gemma 3 has a large, 128K context window, multilingual support in over 140 languages, and is available in more sizes than previous versions. Gemma 3 models are well-suited for a variety of text generation and image understanding tasks, including question answering, summarization, and reasoning. Their relatively small size makes it possible to deploy them in environments with limited resources such as laptops, desktops or your own cloud infrastructure, democratizing access to state of the art AI models and helping foster innovation for everyone. These models were trained on a dataset of text data that includes a wide variety of sources. The 27B model was trained with 14 trillion tokens, the 12B model was trained with 12 trillion tokens, 4B model was trained with 4 trillion tokens and 1B with 2 trillion tokens. Here are the key components: The combination of these diverse data sources is crucial for training a powerful multimodal model that can handle a wide variety of different tasks and data formats.