LMSYS releases open-source chat model Vicuna-13B
What happened
Reported on October 4, LMSYS released Vicuna-13B, an open-source chat model built on LLaMA and fine-tuned on about 70,000 user conversations collected from ShareGPT, at a training cost of roughly $300. In a preliminary evaluation judged by GPT-4, Vicuna-13B reached over 90% of the quality of ChatGPT and Bard, and beat LLaMA and Stanford Alpaca in more than 90% of cases. Code, model weights and an online demo are public, for non-commercial use only.
Written by AI from the coverage · updated 2 hours ago
Coverage
Follow the reports to see the story from different sides.
- AI HOT · ModelsLMSYS 发布开源聊天模型 Vicuna-13B，用 ShareGPT 对话微调 LLaMA，训练成本约 $300
LMSYS 发布开源聊天模型 Vicuna-13B，基于 LLaMA 用 ShareGPT 收集的约 70K 用户对话微调，训练成本约 $300。以 GPT-4 为评判的初步评测显示，Vicuna-13B 达到 ChatGPT 和 Bard 90% 以上的质量，并在超过 90% 的情况下优于 LLaMA 和 Stanford Alpaca。代码、权重和在线 demo 已公开，仅限非商业用途。
Heat over time
Not enough continuous observations to draw a trend yet.