b7c8eecca7ab0493410a288dced2defb4eee6d3bAuthor Julien Chaumond <julien@huggingface.co> Committer GitHub <noreply@github.com> Date 2026-05-18 11:38 UTC Message Use autoload=true for the single /props call (#1)
Previously we called /props with autoload=false and only retried with
autoload=true on 400/404. The fallback was the only path that ever
succeeded when the model wasn't already loaded, so collapse to a single
autoload=true call.
Co-authored-by: julien-agent <Agents+cyolo@huggingface.co> Diff 1 diff --git a/index.ts b/index.ts
2 index 810cf748099a07deda55979da499586c90e249f7..372a02715726f0479d66e7c604afe37fc24ac6d9 100644
3 --- a/index.ts
4 +++ b/index.ts
5 @@ -149,13 +149,10 @@ export default async function (pi: ExtensionAPI) {
6 pendingContext.add(modelId);
7 const controller = new AbortController();
8 const timer = setTimeout(() => controller.abort(), PROPS_TIMEOUT_MS);
9 - const propsBase = `${baseUrl.replace(/\/v1$/, "")}/props?model=${encodeURIComponent(modelId)}&autoload=`;
10 + const propsUrl = `${baseUrl.replace(/\/v1$/, "")}/props?model=${encodeURIComponent(modelId)}&autoload=true`;
11
12 try {
13 - let response = await fetch(`${propsBase}false`, { signal: controller.signal });
14 - if (response.status === 400 || response.status === 404) {
15 - response = await fetch(`${propsBase}true`, { signal: controller.signal });
16 - }
17 + const response = await fetch(propsUrl, { signal: controller.signal });
18 if (!response.ok) {
19 ctx.ui.notify(`[llama-cpp] /props for ${modelId} returned ${response.status}`, "error");
20 return;