Repository navigation
Replies: 1 comment 1 reply
|
这个不是 bug,是凭据引用同名的必然结果。原因一句话就能说清:你手工加的那条路由派生的引用名,和内置那条用的是同一个字符串。 为什么一定撞手工卡片的引用名是按路由名算出来的。 export function deriveKeyRef(provider: string): string {
return `${provider.toUpperCase().replace(/[^A-Z0-9]+/g, '_')}_API_KEY`
}你那条路由叫 再看内置那条: 两边逐字同名。 而 有一处你可能没意识到内置那条路由的名字其实不叫 历史上 立刻能用的两种改法改引用名:把那条 provider profile 的 改路由名:把它改成例如 (顺带说一句, 一点建议我不建议按"每个 provider 应该有独立 key"来提,因为同一个端点配多条路由、共用一把 key 是正当用法,强制隔离反而会挡掉它。 如果你想提改进,可以考虑这个小点:手工卡片在派生出的引用名与已有路由重名时给个提示,或者自动去重( 另外文档侧也值得补一句:Models 页现在没有说明"派生引用 = 边界我核的是 你写的版本 |
Uh oh!
There was an error while loading. Please reload this page.
Environment:
dsh --version→0.2.1-alpha.2Summary:
When adding a model provider and selecting DeepSeek, the manually added DeepSeek model shares the same API key as the built-in DeepSeek model.
Steps to reproduce:
dsh --version→0.2.1-alpha.2Expected:
The manually added DeepSeek model and the built-in one should each have their own independent API key.
Actual:
They share one API key. In practice they're used interchangeably, so editing one also changes the other — meaning one of the two models is always unusable.
Impact:
Can't use both the built-in and the manually added DeepSeek model at the same time.
All reactions