Shenzhen Yuanxiang Information Technology Co., Ltd. recently announced the successful release of China’s largest Mixture of Experts (MoE) open source large model — XVERSE-MoE-A36B. The launch of this model marks a significant advancement in China’s AI field, elevating domestic open source technology to an internationally leading level.

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.

LG's AI research department has launched Exaone 3.0, South Korea's first open-source AI model, which features 7.8 billion parameters. The model aims to accelerate AI research and build a robust AI ecosystem. It performs exceptionally well in Korean and English tasks. Through its open-source strategy, LG demonstrates its technological prowess and may create new revenue streams in cloud computing and AI services. To compete in the global AI landscape, LG is similar to companies like Alibaba, driving cloud business growth through open-source AI models.

CatVTON is a lightweight AI virtual dressing model designed for fashion enthusiasts, with a total parameter count of 899.06M and only 49.57M trainable parameters needed during training. It requires less than 8G of GPU memory for inference and supports 1024x768 high resolution, making it suitable for personal computers. The model can be quickly deployed on ComfyUI and Gradio applications, supporting inference on VITON-HD and DressCode datasets wit....

Mistral AI has launched its flagship model Mistral Large2 today, which boasts 123 billion parameters and a 128k context window. It supports multiple languages, including French, German, and Spanish, as well as programming languages. Mistral Large2 is designed for single-node inference, suitable for long context applications, and performs exceptionally well, achieving an accuracy of 84.0% in the MMLU benchmark test. Additionally, Mistral Large2 has multi-language processing capabilities.

In the realm of artificial intelligence, the competition between open-source and closed-source models has never ceased. The recent release of Meta AI's Llama 3.1 model, however, seems to have drawn a watershed line in this contest. This is not merely the launch of a new model; it is a sign of the maturation of open-source AI, heralding the arrival of a new era.Llama 3.1, developed by Meta AI's team, is a new generation of large-scale language models. In over 150 benchmark tests, its 405B paramet

In the realm of artificial intelligence, Large Language Models (LLMs) are renowned for their exceptional natural language processing capabilities. However, deploying these models in practical applications faces significant challenges due to their high computational costs and memory footprint during the inference phase. To address this issue, researchers have been exploring ways to enhance the efficiency of LLMs. Recently, a method called Q-Sparse has garnered widespread attention.Q-Sparse is a s

Recently, Hugging Face introduced a brand-new AI tool—SmolLM. These are a series of high-performance small language models with parameters ranging from 135M to 1.7B, specifically designed for various devices and applications. Imagine these compact models running efficiently on smartphones and laptops—it's pretty cool!The standout feature of SmolLM models is their small yet powerful nature. They perform exceptionally well with fewer computational resources, helping users protect their privacy.

Meta AI researchers have introduced MobileLLM, a new approach to designing efficient language models for smartphones and other resource-constrained devices. This research was released on June 27, 2024, challenging the assumptions about the necessary scale for effective AI models. The research team consists of members from Meta Reality Labs, PyTorch, and Meta AI Research (FAIR), focusing on optimizing models with fewer than 1 billion parameters. This is a fraction of models like GPT-4, which have

Google has launched a new series of open-source models called 'Gemma'. A comparison between Gemma, Llama 2, and Mistral. The design principles and features of Gemma. Characteristics of the GeGLU activation function used in Gemma.

Google has released the base world model Genie, which features 11 billion parameters and can generate interactive virtual environments. Genie trains without action labels, autonomously identifying different action features and patterns in videos. Genie can learn fine-grained control and simulate a variety of potential actions. It generates video in a cartoon style and can simulate robotic movements and deformable objects, leaning more towards animated graphics compared to Sora.

Orion Star has released the Orion-14B large model, which boasts a parameter size of 14 billion, supports common languages and specialized terminology for ultra-long texts, and achieves an inference speed of 31 tokens/s. It demonstrates excellent multilingual capabilities, particularly excelling in Japanese and Korean. The model offers fine-tuning options and various applications to help enterprises enhance efficiency and decision-making.

["Amazon is training a large language model codenamed Olympus, with an astonishing 2 trillion parameters.", "In contrast, OpenAI's GPT-4 has 1 trillion parameters, making Olympus's parameters twice as many.", "Amazon has strong cloud computing resources and is expected to make waves in the AI field with Olympus.", "The number of parameters is not the only criterion for determining the strength of an AI model; the model's design and training data are equally important.", "It is still unclear when Amazon will publicly release Olympus."]

The Tsinghua-based startup Shuimu Molecular has released ChatDD (Drug Design), a next-generation drug development assistant. ChatDD covers all stages of drug project initiation, preclinical research, and clinical trials, improving the efficiency of drug development. The team also launched ChatDD-FM 100B, the world's first multimodal biomedical dialogue model with 100 billion parameters. ChatDD-FM 100B has excelled in four specialized tasks in the medical field, being the only model with an average score exceeding 90. This innovation is expected to accelerate

ELYZA has released the Japanese LLM 'ELYZA-japanese-Llama-2-7b' based on Meta's Llama 2, with a parameter count reaching 7 billion, performance comparable to GPT-3.5. The model has undergone additional pre-training and unique post-training, achieving the highest scores in a 5-level manual evaluation. Although it has not yet reached the level of closed LLMs, it is already on par with GPT-3.5. ELYZA has successfully developed LLMs in other languages, including English.

Jais is the first open-source large model for Arabic, boasting 13 billion parameters. It performs on par with commercial English ChatGPT models in Arabic tasks. Jais is trained using Cerebras' specialized AI chips, showcasing technological innovation.

The first domestic large model specifically designed for mathematics, MathGPT, with hundreds of billions of parameters, is officially launched. It surpasses GPT-4 in multiple benchmark tests, setting new SOTA records. MathGPT's mathematical computation capabilities cover math problems from elementary school to high school, including various types such as computation problems, application problems, and algebra problems. MathGPT can achieve higher accuracy in solving problems and can explain answers more clearly and communicate them more effectively, better meeting the core needs of users who use AI products to answer math questions.

Xiaomi has released the large-scale pre-trained language model MiLM-6B, which has 6.4 billion parameters. MiLM-6B has achieved the best performance in its size on both C-Eval and CMMLU. MiLM-6B performs excellently and is suitable for various subjects and difficulty levels.