AI 眼镜离线也能识图问答:高通携手 PrismML 发布 1-bit Bonsai,内存占用降至 1/4AI Glasses Can Recognize Images and Answer Questions Offline: Qualcomm Teams Up with PrismML to Launch 1-bit Bonsai, Memory Usage Reduced to 1/4


AI free token/AI免费积分

IT之家 9 月 24 日消息,在 2026 骁龙峰会上,高通携手 PrismML 公司,在骁龙 AR1 Gen 1 智能眼镜平台上,成功本地运行 20 亿参数的 1-bit 量化 Bonsai 视觉模型。

Bonsai 1-bit 模型基于 PrismML 此前发布的 Bonsai 1.7B 版本构建,降低模型存储与推理所需的内存资源,模型精度压缩到 1-bit。官方表示在相同内存容量下可容纳约 4 倍参数,内存占用约为传统 4-bit 模型的 1/4,文本 Token 生成速度翻倍。

AI 眼镜离线也能识图问答:高通携手 PrismML 发布 1-bit Bonsai,内存占用降至 1/4

在落地情形维度,IT之家援引博文介绍,官方表示在识别眼前物体、翻译文字、回答环境相关的难题、给出免手操作的语音协助等情形下,AI 交互不再需要借助连接云端等待,可以直接本地达成交互。

在电池续航、发热和佩戴维度,1-bit 模型可以降低内存和功耗,从而更长电池续航、更少发烫,从而改善佩戴感受。

2026 高通骁龙峰会专题

本文来源:https://www.ithome.com/1/006/772.htm
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

IT Home News, September 24: At the 2026 Snapdragon Summit, Qualcomm, together with PrismML, successfully ran a 2-billion-parameter 1-bit quantized Bonsai vision model locally on the Snapdragon AR1 Gen 1 smart glasses platform.

The Bonsai 1-bit model is built on PrismML's previously released Bonsai 1.7B version, reducing the memory resources required for model storage and inference, with model precision compressed to 1-bit. The company stated that under the same memory capacity, it can accommodate approximately 4 times the parameters, with memory usage at about 1/4 of traditional 4-bit models, and text Token generation speed doubled.

AI Glasses Can Recognize Images and Answer Questions Offline: Qualcomm Teams Up with PrismML to Launch 1-bit Bonsai, Memory Usage Reduced to 1/4

In terms of real-world application scenarios, IT Home, citing the blog post, reports that the company stated that in scenarios such as recognizing objects in front of you, translating text, answering questions related to the surrounding environment, and providing hands-free voice assistance, AI interactions no longer need to connect to the cloud and wait; interactions can be completed directly on-device.

In terms of battery life, heat generation, and wearing comfort, the 1-bit model reduces memory and power consumption, resulting in longer battery life and less heating, thereby improving the wearing experience.

2026 Qualcomm Snapdragon Summit Special Coverage

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry