Search AI Products and News

Researchers at the Hebrew University of Jerusalem recently discovered that in Retrieval Augmented Generation (RAG) systems, the number of documents processed significantly impacts language model performance, even when the total text length remains constant. The research team conducted experiments using 2,417 questions from the MuSiQue validation dataset, each linked to 20 Wikipedia paragraphs. Two to four paragraphs contained relevant answer information, with the remaining paragraphs serving as distractors. To study the impact of the number of documents, the team created multiple data partitions, progressively reducing the number of documents from 20 to...

Recently, research teams from Peking University announced the release of an open-source multimodal model called LLaVA-o1, which is claimed to be the first visual language model capable of spontaneous and systematic reasoning, comparable to GPT-o1. The model excels in six challenging multimodal benchmark tests, with its 11B parameter version outperforming competitors such as Gemini-1.5-pro, GPT-4o-mini, and Llama-3.2-90B-Vision-Instruct.

Kunlun Wanwei Technology Co., Ltd. recently announced that its two new reward models, Skywork-Reward-Gemma-2-27B and Skywork-Reward-Llama-3.1-8B, have performed exceptionally well on the internationally recognized reward model evaluation benchmark RewardBench. Notably, the Skywork-Reward-Gemma-2-27B model achieved the top position and received high recognition from the RewardBench officials.

Arcee AI introduces SuperNova, a 70 billion parameter large language model designed for enterprise deployment, aimed at providing a powerful, owned, data-privacy-focused, stable, and customizable AI solution. SuperNova utilizes Meta's Llama-3.1-70B-Instruct architecture and features advanced instruction-following capabilities through an innovative post-training process. It supports full customization for enterprises and allows deployment in their own cloud environments, ensuring data security.

Nvidia's research team has successfully launched Llama-3.1-Minitron4B using model pruning and distillation techniques. This is a compressed version of the Llama3 model, aimed at implementing artificial intelligence on devices. The model reduces the parameter count of the original 8B model through deep and width pruning techniques while maintaining performance close to larger models. Despite a significant reduction in training data (by 40 times), the model achieved a 16% performance improvement on the MMLU benchmark.

aiOS is the world's first decentralized AI network developed by the "hyperspaceai" organization, based on the Mistral7B model. It aims to fundamentally change the accessibility of artificial intelligence, allowing users to experience cutting-edge decentralized AI computing. Currently, the application is in its early development stage and is available for download and experience by Windows, Linux, and macOS users. It already supports Llama-3, offering users a free trial.According to the official

After months of testing during the Indian elections, Meta has decided to promote its Llama-3 technology-driven AI chatbot to all users in India. However, Meta AI currently supports only English and does not support other local languages.

Meta has released two 24K H100 GPU clusters for training Llama-3. Llama-3 utilizes RoCEv2 networking and Tectonic/Hammerspace's NFS/FUSE network storage. It is expected that Llama-3 will go live by the end of April or mid-May, possibly as a multimodal model and will continue to be open-sourced. Meta plans to have 600,000 H100 computing power by the end of 2024.

Ant Group has released the 4bits quantized version of CodeFuse-CodeLlama-34B. The quantized model can be loaded on a single A10 or RTX4090. The quantized model achieves a performance of 73.8% on the Humaneval pass@1 metric. CodeFuse is Ant's self-developed large model for code generation, aimed at enhancing developers' efficiency by providing intelligent suggestions and real-time support.