Skip to content

armgit / laya

Public Imported from GitHub
0

Non-autoregressive System 1 decision engine. Typed choice, score and yes/no decisions over any text in a single forward pass, in 100+ languages, with a router that picks the right checkpoint per request.

fix(ts): re-read encoder bytes on WebGPU fallback instead of pinning them

nnlgsakibbeb7131fd048
diff --git a/laya-ts/src/providers.ts b/laya-ts/src/providers.ts
index 5b788845..474afb05 100644
--- a/laya-ts/src/providers.ts
+++ b/laya-ts/src/providers.ts
@@ -530,9 +530,17 @@ export async function createWebProvider(
const base = modelUrl.replace(/\/+$/, "");
const encUrl = `${base}/encoder.onnx`;
const headUrl = `${base}/head.onnx`;
+ // (Re-)read through fetchArrayBuffer so repeat reads hit CacheStorage.
+ // Nothing pin-worthy is retained: after each create, buffers are droppable.
+ const readEncoderParts = async () => {
+ const buf = await fetchArrayBuffer(encUrl);
+ const sidecar = await fetchSidecar(encUrl);
+ return { buf, extra: sidecar ? { externalData: [sidecar] } : {} };
+ };
let encBuf: ArrayBuffer;
+ let encExtra: Record<string, unknown>;
try {
- encBuf = await fetchArrayBuffer(encUrl);
+ ({ buf: encBuf, extra: encExtra } = await readEncoderParts());
} catch {
throw new Error(`Incompatible model: 'encoder.onnx' not found (expected ${encUrl}).`);
}
@@ -549,9 +557,7 @@ export async function createWebProvider(
}
// Split ONNX references its weights relatively ("encoder.onnx.data"); buffered
// sessions have no filesystem, so mount the sidecar via externalData.
- const encSidecar = await fetchSidecar(encUrl);
const headSidecar = await fetchSidecar(headUrl);
- const encExtra = encSidecar ? { externalData: [encSidecar] } : {};
const headExtra = headSidecar ? { externalData: [headSidecar] } : {};
let enc: any;
try {
@@ -599,10 +605,19 @@ export async function createWebProvider(
console.warn(
`laya: WebGPU encoder run failed (${String((e as Error)?.message ?? e)}); falling back to WASM.`,
);
- encWasm ??= await ort.InferenceSession.create(new Uint8Array(encBuf), {
- executionProviders: ["wasm"],
- ...encExtra,
- });
+ if (!encWasm) {
+ // Re-read through the cache instead of pinning 1GB+ for the agent's life.
+ let parts;
+ try {
+ parts = await readEncoderParts();
+ } catch {
+ throw e;
+ }
+ encWasm = await ort.InferenceSession.create(new Uint8Array(parts.buf), {
+ executionProviders: ["wasm"],
+ ...parts.extra,
+ });
+ }
try {
return await runEncoderOn(encWasm, b);
} catch (e2) {