dgf1988/codewhale
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

fix: vLLM provider — pass through reasoning_effort, downgrade max to high (#2169) (#2170)

Browse Source
This commit is contained in:
Hanmiao Li
2026-05-26 23:30:47 +08:00
committed by GitHub
parent d107b7468d
commit 81480ba099
1 changed files with 11 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
crates/tui/src/client.rs
+11 -2
View File
@@ -937,7 +937,14 @@ pub(super) fn apply_reasoning_effort(
body["chat_template_kwargs"] = json!({
"enable_thinking": true,
});
body["reasoning_effort"] = json!("high");
// vLLM supports low/medium/high natively — pass through the
// user-chosen value instead of hard-coding "high".
let value = match normalized.as_str() {
"low" | "minimal" => "low",
"medium" | "mid" => "medium",
_ => "high",
};
body["reasoning_effort"] = json!(value);
}
ApiProvider::Openai
| ApiProvider::Atlascloud
@@ -967,7 +974,9 @@ pub(super) fn apply_reasoning_effort(
body["chat_template_kwargs"] = json!({
"enable_thinking": true,
});
body["reasoning_effort"] = json!("max");
// vLLM only supports none/low/medium/high — downgrade
// "max" to "high" instead of sending an invalid value.
body["reasoning_effort"] = json!("high");
}
ApiProvider::Openai
| ApiProvider::Atlascloud