mirror of
https://github.com/diegosouzapw/OmniRoute.git
synced 2026-08-26 09:02:11 +03:00
cd091ab878
The model ships only as `qwen3.8-max-preview` across every provider that serves it (bailian-coding-plan, qoder, qwen-cloud-token-plan, qwen-web), so the bare `qwen3.8-max` missed MODEL_SPECS: the chatCore context preflight fell back to contextManager's `default: 128000` and rejected prompts with `context_length_exceeded` despite the model's real 1M window, and the unknown id would have reached the upstream verbatim.
Both symptoms share one cause, so the alias goes in BUILT_IN_ALIASES, which resolveLifecycle() applies before the preflight and before dispatch.
Merged with the inherited OmniGlyph base-red (#9985) documented: its two failing compression tests were reproduced on the pure base tip aa912c42a7, with no commit from this branch.
55 lines
2.7 KiB
TypeScript
55 lines
2.7 KiB
TypeScript
import assert from "node:assert/strict";
|
|
import { test } from "node:test";
|
|
import { MODEL_SPECS } from "../../src/shared/constants/modelSpecs.ts";
|
|
import { resolveModelAlias } from "../../open-sse/services/modelDeprecation.ts";
|
|
import { resolveLifecycle } from "../../open-sse/handlers/chatCore/modelLifecyclePolicy.ts";
|
|
|
|
/**
|
|
* Bare `qwen3.8-max` was an unroutable id: the model ships everywhere as
|
|
* `qwen3.8-max-preview` (bailian-coding-plan, qoder, qwen-cloud-token-plan, qwen-web),
|
|
* and nothing in the repo declared the short form. A client sending it therefore
|
|
*
|
|
* 1. missed MODEL_SPECS, so `getModelContextLimit()` fell through to the
|
|
* `default: 128000` in open-sse/services/contextManager.ts, and the chatCore
|
|
* preflight rejected any prompt above 128k with `context_length_exceeded`
|
|
* ("Input exceeds context window ... limit 128000") even though the real
|
|
* window is 1M; and
|
|
* 2. would have been dispatched verbatim to the upstream, which only knows the
|
|
* `-preview` id.
|
|
*
|
|
* Both symptoms have one cause — the missing id — so the fix belongs in the
|
|
* deprecation/rename alias map (`BUILT_IN_ALIASES`), which `resolveLifecycle()`
|
|
* applies at open-sse/handlers/chatCore.ts:755, well before both the context
|
|
* preflight and the upstream dispatch. A MODEL_SPECS `aliases` entry would have
|
|
* fixed only (1): spec aliases resolve capabilities, never the dispatched id.
|
|
*/
|
|
|
|
const BARE = "qwen3.8-max";
|
|
const CANONICAL = "qwen3.8-max-preview";
|
|
|
|
test("bare qwen3.8-max resolves to the canonical -preview id", () => {
|
|
assert.equal(resolveModelAlias(BARE), CANONICAL);
|
|
});
|
|
|
|
test("the canonical id is a no-op through the alias map (no double rewrite)", () => {
|
|
assert.equal(resolveModelAlias(CANONICAL), CANONICAL);
|
|
});
|
|
|
|
test("the alias target carries the real 1M window, not the 128k fallback", () => {
|
|
const spec = MODEL_SPECS[CANONICAL];
|
|
assert.ok(spec, `MODEL_SPECS is missing ${CANONICAL}`);
|
|
assert.equal(spec.contextWindow, 1_000_000);
|
|
// The bare id must NOT gain its own spec entry — a second source of truth for the
|
|
// same model is what lets the two ids drift apart again.
|
|
assert.equal(MODEL_SPECS[BARE], undefined);
|
|
});
|
|
|
|
test("chatCore lifecycle resolution rewrites the model before dispatch", () => {
|
|
for (const provider of ["qwen-cloud-token-plan", "qoder", "bailian-coding-plan", "qwen-web"]) {
|
|
const [resolvedModel, effectiveModel, lifecycleError] = resolveLifecycle(provider, BARE);
|
|
assert.equal(resolvedModel, CANONICAL, `resolvedModel for ${provider}`);
|
|
assert.equal(effectiveModel, CANONICAL, `effectiveModel for ${provider}`);
|
|
assert.equal(lifecycleError, null, `unexpected lifecycle rejection for ${provider}`);
|
|
}
|
|
});
|