1. update nanokvm cua desc

This commit is contained in:
zepan
2025-10-11 11:27:48 +08:00
parent d5cfcf40ba
commit 96a6cd0439
2 changed files with 2 additions and 0 deletions

View File

@@ -179,6 +179,7 @@ As of October 2025, the open-source **Qwen3-VL** models have significantly impro
- `qwen3-vl-235b-a22b-instruct` surpasses last years `qwen-vl-max`
- `qwen3-vl-30b-a3b-instruct` outperforms `qwen2.5-vl-72b-instruct`
Both now meet the threshold for basic computer control tasks.
> It is expected that Qwen3-VL-4B will be released in mid-October. It may also feature CUA capabilities, which would make self-deployment on regular consumer-grade computers possible!
However:
- The **235B** model requires at least **4×H100 GPUs** (320GB total)—impractical for most users.

View File

@@ -140,6 +140,7 @@ CUA网页是电脑手机浏览器兼容的布局电脑上的页面布局
### 硬件配置
得益于Qwen3-VL系列的发布用户自部署VLM服务实现CUA功能也成为了现实。
在2025年10月最新发布的Qwen3-VL系列开源模型的能力大幅提高qwen3-vl-235b-a22b-instruct能力超越了去年的qwen-vl-max, qwen3-vl-30b-a3b-instruct超越了qwen2.5-vl-72b-instruct, 都达到了完成基础电脑操作的能力门槛。
> 预计在10月中将会发布Qwen3-VL-4B, 也许可能也具备CUA能力如果可以的话将可以在普通消费级电脑上自部署
qwen3-vl-235b-a22b-instruct是较大模型至少需要4xH100 (4x80=320GB) 来运行FP8模型对于普通用户来说比较困难。
我们主要介绍 qwen3-vl-30b-a3b-instruct 的自部署演示。