DSH | dsh-vision-autoswitch | 有图自动切Vision,无图回落到Pro/Flash Image in → Vision, image out → back to Pro/Flash #5453
CultOfLuna
started this conversation in
Show Your Plugins!
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
项目地址 / Project URL:
https://github.com/CultOfLuna/dsh-vision-autoswitch
npm 包:
dsh-vision-autoswitch(已发布 v0.1.2)项目介绍 / Introduction:
当你的对话里出现图片时,插件会在该回合临时切换到视觉模型
deepseek-v4-flash-vision-exp处理;回合结束后自动回落到你之前选择的模型(Flash / Pro)。不用手动切模型、不依赖下一条输入再识别,长对话也不会失灵;模型栏所见即所得,实际模型与 UI 显示始终一致。When an image appears in your conversation, this plugin temporarily routes that turn to the vision model
deepseek-v4-flash-vision-exp, then automatically falls back to your previously selected model (Flash/Pro) when the turn ends. No manual switching, no waiting for the next message, and it stays correct across long conversations — the model bar always reflects the model actually in use.与 DSH 的集成方式 / How it integrates with DSH:
agent/request阶段检测当前回合是否含图(contentHasImage),含图则把 provider/model 改写为视觉模型;session/event的turn/end,回合结束后写回model/selection恢复原模型;llm.resolveModelInfo为视觉模型声明image输入能力,确保图片请求通过 DSH 的准入校验。安装(npm + 在 profile 的
cordis.patch.yml追加):截图 / Screenshots:
All reactions