全站搜索

AI 模型 2119

• gemma-4-E2B-it-RKLLM↗

  HanzoHuang

  Text generation, image questions, and short audio transcription on Rockchip RK3576 and RK3588, converted from google/gemma-4-E2B-it. **RK3576 preview:** image and audio requests work in the functional tests, but the vision encoder misses the numerical accuracy target. RK3576 W8A8 also fails one exact-recall check. See the measured results below before choosing this variant. RK3588 W8A8 passes the functional and encoder checks performed here. The new matched multimodal set is under **`v1.3.1/`**. Use the exact SoC for your board. Each variant needs its language file and vision encoder; add the audio encoder to enable speech input. The two RK3576 variants share the same encoders.

• align-base↗

  kakaobrain

  The ALIGN model was proposed in "Scaling Up Visual and Vision-Language Representation Learning With Noisy Text Supervision" by Chao Jia, Yinfei Yang, Ye Xia, Yi-Ting Chen, Zarana Parekh, Hieu Pham, Quoc V. Le, Yunhsuan Sung, Zhen Li, Tom Duerig. ALIGN features a dual-encoder architecture with EfficientNet as its vision encoder and BERT as its text encoder, and learns to align visual and text representations with contrastive learning. Unlike previous work, ALIGN leverages a massive noisy dataset and shows that the scale of the corpus can be used to achieve SOTA representations with a simple recipe. The code for ALIGN was not publicly released, the base model is converted from the original implementation of the Kakao Brain team. This implementation follows the same architecture and hyperparameters as provided in the original Google model but is trained on the open source COYO dataset. Google’s ALIGN model, while trained on a huge dataset of 1.8 billion image-text pairs, cannot be replicated as the datasets is not public. Kakao Brain's ALIGN is on-par or outperforms Google ALIGN's reported metrics despite being trained on the much smaller, albeit carefully curated COYO-700M dataset. COYO is available on the hub as a dataset.

• Qwen-Image-2.1-Text-Encoder-Heretic-DHQ-GGUF↗

  pottokao

  **DHQ = Dual-Head Quant.** Same heretic weights as Qwen-Image-2.1-Text-Encoder-Heretic-GGUF, quantized a different way: instead of llama.cpp's fixed rules, every tensor gets the precision that matters to **what the Qwen-Image-2.1 DiT actually reads** (measured per tensor with gradients through the DiT's text input), plus an importance matrix weighted the same way. Each file is **exactly the size of the standard llama.cpp file of the same name**, and better at every level. Tuned for **Qwen-Image-2.1 only**. Other models that use Qwen3-VL-8B (Ideogram 4, Boogu, JoyImageEdit) can load these files, but they were not tuned for them. Need Q5/Q6/Q8? Use the standard files in the GGUF repo — at those sizes the standard recipe is already as good. DHQ files are quantized **for use as a diffusion model's text encoder**: each tensor's precision is chosen by how much its quantization error changes what the target DiT reads (measured through the DiT's own text-input stage), not by next-token loss. They stay normal LLM GGUFs, so the same file can still write or rewrite prompts: where a model is also used as a prompt enhancer (Z-Image-Engineer-V6, the heretic Qwen3-VL-8B), every file is also checked for that (the PE column) and the tables show where it gets worse.

• mms-tts-ara↗

  facebook

  This model is part of Facebook's Massively Multilingual Speech project, aiming to provide speech technology across a diverse range of languages. You can find more details about the supported languages and their ISO 639-3 codes in the MMS Language Coverage Overview, and see all MMS-TTS checkpoints on the Hugging Face Hub: facebook/mms-tts. VITS (**V**ariational **I**nference with adversarial learning for end-to-end **T**ext-to-**S**peech) is an end-to-end speech synthesis model that predicts a speech waveform conditional on an input text sequence. It is a conditional variational autoencoder (VAE) comprised of a posterior encoder, decoder, and conditional prior. A set of spectrogram-based acoustic features are predicted by the flow-based module, which is formed of a Transformer-based text encoder and multiple coupling layers. The spectrogram is decoded using a stack of transposed convolutional layers, much in the same style as the HiFi-GAN vocoder. Motivated by the one-to-many nature of the TTS problem, where the same text input can be spoken in multiple ways, the model also includes a stochastic duration predictor, which allows the model to synthesise speech with different rhythms from the same input text.

• data2vec-audio-base-960h↗

  facebook

  The base model pretrained and fine-tuned on 960 hours of Librispeech on 16kHz sampled speech audio. When using the model make sure that your speech input is also sampled at 16Khz. While the general idea of self-supervised learning is identical across modalities, the actual algorithms and objectives differ widely because they were developed with a single modality in mind. To get us closer to general self-supervised learning, we present data2vec, a framework that uses the same learning method for either speech, NLP or computer vision. The core idea is to predict latent representations of the full input data based on a masked view of the input in a self-distillation setup using a standard Transformer architecture. Instead of predicting modality-specific targets such as words, visual tokens or units of human speech which are local in nature, data2vec predicts contextualized latent representations that contain information from the entire input. Experiments on the major benchmarks of speech recognition, image classification, and natural language understanding demonstrate a new state of the art or competitive performance to predominant approaches. The original model can be found under https://github.com/pytorch/fairseq/tree/main/examples/data2vec .

• Ornith-1.0-9B-gguf↗

  samuelchristlie

• Step-3.7-Flash-GGUF↗

  阶跃星辰

  Step-3.7-Flash is a 198B-parameter sparse Mixture-of-Experts vision-language model from StepFun-ai, activating ~11B parameters per token for up to 400 t/s throughput. It pairs a 196B-parameter language backbone with a 1.8B-parameter vision encoder for native image understanding, supports a 256K context window, and offers three selectable reasoning levels (low / medium / high) to balance speed, cost, and depth. Built for agentic workloads — tool calling, multi-step reasoning, code, and math — with native multilingual coverage. A separate `mmproj` projector ships alongside the language quants for multimodal inference. With 128 GB of unified memory (Mac Studio, DGX Spark, Ryzen AI Max+ 395, etc.), you can privately host Step-3.7-Flash: Q4 quants and below run at full 256K context with high precision. The GGUF quantization files in this repository are derivative works of `stepfun-ai/Step-3.7-Flash` and are released under the same **Apache 2.0** license.

• Sakura-Qwen-Image-2.1-Turbo-Uncensored-GGUF↗

  webmp3

  **Built with Qwen.** Community quantization of Qwen/Qwen-Image-2.1-Turbo (8-step distilled, 7B DiT) with an **abliterated text encoder**. Independent work, not made or endorsed by Alibaba/Qwen. **Non-commercial use only** (Qwen RESEARCH LICENSE AGREEMENT, copy in `LICENSE`). For adults only; you are responsible for what you generate and for the laws that apply to you. Q8_0 image model and our 8bit uncensored text encoder, stable-diffusion.cpp (Vulkan, Radeon 8060S), 1024x1024, 8 steps, CFG 1.0, `euler`, official Turbo sigmas. One seed per prompt and no re-rolls; the prompts themselves were chosen from a larger set of candidates. About 77 s per image including model loading; the GPU memory in use on our machine peaked at 18.3 GiB while generating (Windows counter for the whole GPU, 1.5 GiB of it were in use before the run). The images were made with stable-diffusion.cpp, not with ComfyUI. The prompts are neutral. These images say nothing about what the ablated text encoder changes for restricted content; they only show that ordinary prompts still work. The same nine prompts and seeds are used in the sibling repository Sakura-Qwen-Image-2.1-Turbo-GGUF with the official encoder, so the two sets can be compared.

• InternVL2_5-1B↗

  OpenGVLab

  We are excited to introduce **InternVL 2.5**, an advanced multimodal large language model (MLLM) series that builds upon InternVL 2.0, maintaining its core model architecture while introducing significant enhancements in training and testing strategies as well as data quality. As shown in the following figure, InternVL 2.5 retains the same model architecture as its predecessors, InternVL 1.5 and 2.0, following the "ViT-MLP-LLM" paradigm. In this new version, we integrate a newly incrementally pre-trained InternViT with various pre-trained LLMs, including InternLM 2.5 and Qwen 2.5, using a randomly initialized MLP projector. As in the previous version, we applied a pixel unshuffle operation, reducing the number of visual tokens to one-quarter of the original. Besides, we adopted a similar dynamic resolution strategy as InternVL 1.5, dividing images into tiles of 448×448 pixels. The key difference, starting from InternVL 2.0, is that we additionally introduced support for multi-image and video data.

• mustango↗

  declare-lab

  Meet Mustango, an exciting addition to the vibrant landscape of Multimodal Large Language Models designed for controlled music generation. Mustango leverages Latent Diffusion Model (LDM), Flan-T5, and musical features to do the magic! We use the `accelerate` package from Hugging Face for multi-gpu training. Run `accelerate config` from terminal and set up your run configuration by the answering the questions asked. The `--model_type` flag allows to choose either Mustango, or Tango to be trained with the same code. However, do note that you also need to change `--unet_model_config` to the relevant config: diffusion_model_config_munet for Mustango; diffusion_model_config for Tango.