Reasoning intensity is not selectable for third-party models.第三方模型无法选择推理强度 #122
Replies: 3 comments 2 replies
|
Confirmed from the current source and an actual session log: reasoning effort is an exact-model capability, not a provider-wide UI option. A custom route normally has no catalog capability to inherit, so declare llm-pi-ai:
providers:
my-gateway:
apiKeyEnv: MY_GATEWAY_API_KEY
api: openai-responses
baseURL: https://gateway.example/v1
models:
- id: my-reasoning-model
name: My Reasoning Model
input: [text, image]
reasoningEfforts:
low: low
medium: medium
high: high
xhigh: xhigh
max: maxThe left side is the DSH level ID; the right side is the wire value sent upstream. If the gateway expects different spellings, mappings such as After refreshing and selecting High, a new request was verified in the session log as: {"provider":"my-gateway","model":"my-reasoning-model","reasoningEffort":"high"}
Note: 中文总结:第三方模型需要在具体模型条目中声明 |
|
有用!太厉害了 |
|
同感,已解决:第三方(llm-pi-ai 手工声明)模型没有「推理等级」,是因为适配器默认不暴露 reasoningEfforts,需要每个模型声明档位。 我做的插件 dsh-thinking-effort(MIT):
dsh plugin --profile <profile> add github:hytime/dsh-thinking-effort说明与安装:https://github.com/hytime/dsh-thinking-effort/blob/main/INSTALL.md · 宣传帖:#1817 English: built dsh-thinking-effort — host auto-declares default effort levels for hand-declared pi-ai models, and a Settings page lets you pick levels per model with custom wire values. |

Uh oh!
There was an error while loading. Please reload this page.
暂时没看到issue page,先发这里了
All reactions