Compare commits

...

3 Commits

Author SHA1 Message Date
Diego Rodrigues de Sa e Souza
804b085d36 Merge branch 'release/v3.8.55' into feat/paralon-cloud-extras 2026-11-20 19:12:57 -03:00
diegosouzapw
a15f362b5c docs(changelog): add the ParalonCloud extras fragment for #20794
Co-authored-by: Robert Jitaru <42837438+robertjitaru@users.noreply.github.com>
2026-11-20 13:13:56 -03:00
diegosouzapw
53df4e8278 feat(providers): ParalonCloud passthrough models, context lengths and Responses URL
ParalonCloud landed in #18356 with a two-model seed and live discovery. This
carries over what #16587 still adds on top of it:

- passthroughModels on both the registry entry and the API-key catalog entry,
  so any model id the live /v1/models list returns is accepted (the catalog
  follows the GPUs online on the network)
- per-model context lengths on the offline seed (qwen3.8-27b 131072,
  qwen3-3b 32768) and the qwen3-3b display name matching the live catalog
  (Qwen3 4B)
- the native /v1/responses endpoint recorded as responsesBaseUrl
- dashboard authHint (prlc_ keys) and apiHint (base URL, prepaid credits)

Tests: provider-paraloncloud-registry gains passthrough, context-length,
Responses-URL and not-an-aggregator cases (2 fail on the base, 8/8 pass here).

Carried over from #16587.

Co-authored-by: Robert Jitaru <42837438+robertjitaru@users.noreply.github.com>
2026-11-20 13:00:50 -03:00
4 changed files with 43 additions and 3 deletions

View File

@@ -0,0 +1 @@
- **feat(providers):** ParalonCloud (`paraloncloud`) now accepts any model id its live `/v1/models` catalog returns (`passthroughModels`), seeds its offline models with the catalog's context lengths, records its native `/v1/responses` endpoint and shows key/base-URL hints in the dashboard ([#20794](https://github.com/cheaper-inference/OmniRoute/pull/20794), carried over from [#16587](https://github.com/cheaper-inference/OmniRoute/pull/16587)) — thanks @robertjitaru

View File

@@ -7,15 +7,19 @@ import { buildOpenAiCompatibleRegistryEntry } from "../../shared.ts";
*
* Embeddings (`multilingual-e5-small`) live in embeddingRegistry. `GET /v1/models` is
* key-gated, so `modelsUrl` drives key validation and dashboard discovery; the seed below
* is only the offline fallback.
* is only the offline fallback, with the per-model context maxima the catalog reports.
* The catalog follows the GPUs online on the network, so any model id the live list
* returns is accepted (`passthroughModels`). `POST /v1/responses` is served natively.
*/
export const paraloncloudProvider: RegistryEntry = buildOpenAiCompatibleRegistryEntry({
id: "paraloncloud",
alias: "paralon",
baseUrl: "https://paraloncloud.com/v1/chat/completions",
responsesBaseUrl: "https://paraloncloud.com/v1/responses",
modelsUrl: "https://paraloncloud.com/v1/models",
models: [
{ id: "qwen3.8-27b", name: "Qwen3.8 27B" },
{ id: "qwen3-3b", name: "Qwen3 3B" },
{ id: "qwen3.8-27b", name: "Qwen3.8 27B", contextLength: 131072 },
{ id: "qwen3-3b", name: "Qwen3 4B", contextLength: 32768 },
],
passthroughModels: true,
});

View File

@@ -167,7 +167,11 @@ export const APIKEY_PROVIDERS_INFERENCE = {
icon: "cloud",
color: "#2EE6A6",
textIcon: "PC",
passthroughModels: true,
website: "https://paraloncloud.com",
authHint: "Create an API key (prlc_…) in the ParalonCloud console, then paste it here.",
apiHint:
"OpenAI-compatible base URL: https://paraloncloud.com/v1. Prepaid credits, priced per token. The catalog follows the GPUs online on the network; check the live model list before selecting one.",
notice: {
apiKeyUrl: "https://paraloncloud.com/console",
},

View File

@@ -13,6 +13,8 @@ const { getEmbeddingProvider, getEmbeddingDimension, parseEmbeddingModel } =
const { isNamedOpenAIStyleProvider } =
await import("../../src/app/api/providers/[id]/models/discovery/providerSets.ts");
const { DefaultExecutor, getExecutor } = await import("../../open-sse/executors/index.ts");
const { isValidModel } = await import("../../src/shared/constants/models.ts");
const { AGGREGATOR_PROVIDER_IDS } = await import("../../src/shared/constants/providers.ts");
const PARALON_API = "https://paraloncloud.com/v1";
@@ -73,3 +75,32 @@ test("paraloncloud resolves in the embedding registry", () => {
model: "multilingual-e5-small",
});
});
// #16587 (robertjitaru): the ParalonCloud catalog follows the GPUs online on the network, so
// any model id the live /v1/models list returns must be accepted (passthrough), and the
// offline seed carries the context lengths the catalog reports.
test("paraloncloud accepts any model the live catalog returns (passthrough)", () => {
assert.equal(providerRegistry.paraloncloud.passthroughModels, true);
assert.equal(APIKEY_PROVIDERS.paraloncloud.passthroughModels, true);
assert.equal(isValidModel("paraloncloud", "future/live-catalog-model"), true);
assert.equal(isValidModel("paralon", "future/live-catalog-model"), true);
assert.match(APIKEY_PROVIDERS.paraloncloud.authHint, /prlc_/);
assert.match(APIKEY_PROVIDERS.paraloncloud.apiHint, /live model list/);
});
test("paraloncloud seeds carry context lengths and the Responses endpoint", () => {
const entry = providerRegistry.paraloncloud;
assert.equal(entry.responsesBaseUrl, `${PARALON_API}/responses`);
assert.deepEqual(
entry.models.map((m: { id: string; contextLength?: number }) => [m.id, m.contextLength]),
[
["qwen3.8-27b", 131072],
["qwen3-3b", 32768],
]
);
});
test("paraloncloud is not listed as an aggregator", () => {
// It serves its own models on its own GPU network; it does not route to other APIs.
assert.equal(AGGREGATOR_PROVIDER_IDS.has("paraloncloud"), false);
});