90年代的黄河路

在AI竞赛日益激烈的背景下,法国人工智能初创公司Mistral AI在近期加入竞争行列,推出其旗舰开源模型的下一代产品——Mistral Large 2。这一新模型拥有1230亿个参数,其在代码生成、数学、推理等关键领域的能力与OpenAI、Meta等顶级模型不相上下。Mistral Large 2的发布,使得开源大模型的赛道变得愈发热闹。

Mistral AI的这一新作紧随Meta推出的Llama 3.1之后,参数量虽略低于Llama 3.1的4050亿,但性能表现接近,且在多个基准测试中与GPT-4、Anthropic的Claude 3.5 Sonnet等尖端模型并驾齐驱。Mistral Large 2在性能和服务成本上刷新了新标准,预训练版本在MMLU上的准确率达到了84.0%。同时,Mistral AI在增强模型的推理能力上下足了功夫,减少了模型产生“幻觉”的倾向,确保输出的准确性和可靠性。在数学基准测试中,Mistral Large 2展现了其解决问题能力的显著提升。

Mistral Large 2的一大亮点是其支持数十种语言,包括法语、德语、西班牙语、意大利语、葡萄牙语、阿拉伯语、印地语、俄语、中文、日语和韩语,以及80多种编程语言,如Python、Java、C、C++、JavaScript和Bash等。这使得Mistral Large 2能够广泛应用于不同领域,满足用户在多语言环境下的需求。

值得注意的是,尽管Mistral Large 2是开放的,但仅限于研究和非商业用途,允许第三方根据需求对模型进行微调。对于商业用途,用户需获取Mistral AI的商业许可证。Mistral AI通过持续优化和创新,致力于推动AI技术的发展,为用户提供性能更优、成本效益更高的AI解决方案。

英语如下:

News Title: “Mistral AI Launches 12.3 Billion Parameter Open-Source Model, Challenging Llama 3.1’s Dominance in AI Race”

Keywords: AI competition, open-source model, performance parity

News Content: Amidst intensifying competition in the AI sphere, the French AI startup, MISTRAI AI, has recently entered the fray with its next-generation flagship open-source model, MISTRAI Large 2. This new model boasts 12.3 billion parameters and demonstrates performance on par with industry leaders like OpenAI and Meta, excelling in code generation, mathematics, and inference. The launch of MISTRAI Large 2 has added a new dimension of excitement to the open-source large model landscape.

Following the release of Meta’s Llama 3.1, MISTRAI AI’s MISTRAI Large 2, while having fewer parameters (approximately 40.5 billion fewer than Llama 3.1), matches its performance in multiple benchmark tests, and competes alongside cutting-edge models like GPT-4, Anthropic’s Claude 3.5 Sonnet. The model has set a new benchmark in performance and service costs, achieving an accuracy of 84.0% on the MMLU pre-training version. MISTRAI AI has also focused on enhancing the model’s inference capabilities, reducing its tendency to produce “hallucinations” and ensuring high accuracy and reliability in its outputs. In mathematics benchmarks, MISTRAI Large 2 showcases a significant improvement in its problem-solving capacity.

One of the standout features of MISTRAI Large 2 is its support for over 80 languages, including French, German, Spanish, Italian, Portuguese, Arabic, Hindi, Russian, Chinese, Japanese, and Korean, along with a wide range of programming languages such as Python, Java, C, C++, JavaScript, and Bash. This broad language support enables MISTRAI Large 2 to be widely applicable across various fields and meet the needs of users in multi-language environments.

It is worth noting that, while MISTRAI Large 2 is open-source, it is restricted to research and non-commercial use, allowing third parties to fine-tune the model based on their needs. For commercial purposes, users need to obtain a commercial license from MISTRAI AI. MISTRAI AI is committed to driving advancements in AI technology through continuous optimization and innovation, providing users with more superior and cost-effective AI solutions.

【来源】https://www.jiqizhixin.com/articles/2024-07-25

Views: 3

发表回复

您的邮箱地址不会被公开。 必填项已用 * 标注