近日，Mistral AI 发表了 Mixtral 8x7B 模型的论文，该模型在 MMLU 基准测试中表现优异，领先于 GPT-3.5 和 LLaMA 2 70B。
Mistral AI 在论文中详细描述了 Mixtral 8x7B 模型的架构，并进行了与 LLaMA 2 70B 和 GPT-3.5 的广泛基准测试比较。结果显示，在 MMLU 基准测试中，Mixtral 8x7B 的表现超越了这两个模型。
值得注意的是，Mixtral 8x7B 的表现与更大模型（例如 Gemini Ultra 或 GPT-4）相当，可以达到 85% 到 90% 的水平，具体取决于提示方法。
This news report is about the recent publication of the Mixtral 8x7B model paper by Mistral AI. The model has shown exceptional performance in the MMLU benchmark test, outperforming GPT-3.5 and LLaMA 2 70B.
The paper by Mistral AI provides a detailed description of the architecture of the Mixtral 8x7B model and compares it extensively with LLaMA 2 70B and GPT-3.5 in benchmark tests. The results indicate that the Mixtral 8x7B model has surpassed both of these models in the MMLU benchmark test.
It is worth noting that the performance of the Mixtral 8x7B model is comparable to that of larger models (such as Gemini Ultra or GPT-4), achieving levels of 85% to 90%, depending on the method of prompting.
【来源】https://the-decoder.com/mixtral-8x7b-is-currently-the-best-open-source-llm-surpassing-gpt-3-5/
Views: 1