1. Introduction 2. Pretrain model 3. Using SimeCSE_Vietnamese with `sentences-transformers` - Installation - Example usage 4. Using SimeCSE_Vietnamese with `transformers` - Installation - Example usage Pre-trained SimeCSE_Vietnamese models are the state-of-the-art of Sentence Embeddings with Vietnamese : - SimeCSE_Vietnamese pre-training approach is based on SimCSE which optimizes the SimeCSE_Vietnamese pre-training procedure for more robust performance. - SimeCSE_Vietnamese encode input sentences using a pre-trained language model such as PhoBert - SimeCSE_Vietnamese works with both unlabeled and labeled data.
`voyage-4-nano` is a state-of-the-art text embedding model from the Voyage 4 series, designed for high-performance semantic search and retrieval tasks. This model features: - **Developed by**: Voyage AI - **Supported Language(s)**: Multilingual - **Context Length**: 32000 - **Parameters**: 180M [Non-embedding] + 160M [Embedding] - **License**: Apache 2.0 For detailed performance metrics and benchmarks, please refer to: - 📝 Voyage-4 Release Blog Post - 📊 Evaluation Spreadsheet
This is a sentence-transformers model: It maps sentences & paragraphs to a 384 dimensional dense vector space and can be used for tasks like clustering or semantic search. For an automated evaluation of this model, see the *Sentence Embeddings Benchmark*: https://seb.sbert.net
WeMM-Embedding-2B is a universal multimodal embedding model built on Qwen3.5. It accepts text, images, videos, visual documents, and interleaved multimodal inputs, and returns a 2,048-dimensional L2-normalized embedding. Audio input is not supported. Each input is a string, a URL or path, a `PIL.Image`, or a dict combining `image`, `video`, and `text` keys. Chat messages such as `{"role": "user", "content": [{"type": "image", "image": ...}, {"type": "text", "text": ...}]}` are also accepted, which is the way to interleave several images or videos in one input. Use a dimension listed in `model.config.matryoshka_dimensions`. On MMEB-v2, 256-dimensional embeddings retain 98.7% of the full-dimensional image and video performance.
WeMM-Embedding-9B is a universal multimodal embedding model built on Qwen3.5. It accepts text, images, videos, visual documents, and interleaved multimodal inputs, and returns a 4,096-dimensional L2-normalized embedding. Audio input is not supported. Each input is a string, a URL or path, a `PIL.Image`, or a dict combining `image`, `video`, and `text` keys. Chat messages such as `{"role": "user", "content": [{"type": "image", "image": ...}, {"type": "text", "text": ...}]}` are also accepted, which is the way to interleave several images or videos in one input. Results on 78 datasets from Table 1 of the technical report. Image and video tasks use Hit@1, while visual-document tasks use NDCG@5. Higher is better.
This is the **fine-tuned version** of the EAT-base (Epoch 30, Pre-trained Checkpoint), further trained on the AS-2M dataset. Compared to the pre-trained model, this version provides **enhanced audio representations** and typically yields better performance in downstream audio understanding tasks such as classification and captioning. For more details on the EAT framework, please refer to the GitHub repository and our paper EAT: Self-Supervised Pre-Training with Efficient Audio Transformer. The model supports both **frame-level** (\~50Hz) and **utterance-level** (CLS token) representations. See the feature extraction guide for detailed instructions.
