seanlee97
Professional software vendor delivering innovative solutions on the Softono platform. Specialized in both open-source and proprietary software development.
Software by seanlee97
AnglE
AnglE is a powerful library for training and inferring sentence embeddings, recognized for achieving State-of-the-Art performance on STS and MTEB leaderboards. Designed for flexibility and ease of use, it enables users to train BERT-based models and large language models like LLaMA, Mistral, and Qwen with minimal code. The framework supports diverse architectural backbones, including standard BERT variants, modern LLMs, and bi-directional LLM configurations. It offers multiple advanced training loss functions, such as AnglE loss, Contrastive loss, CoSENT loss, and Espresso loss, to optimize embedding quality for specific tasks. AnglE facilitates efficient training on both single and multi-GPU setups. Pretrained models available through the library cover general-purpose English text, code similarity, and specialized medical domains. Originally presented in the ACL 2024 paper Angle-optimized Text Embeddings, AnglE also integrates research from NAACL 2024 regarding backward dependency enhancement in large langua