Skip to content

fix(llm-pi-ai): 目录路由的模型发现改为合并线上清单 #309

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
FudanSE-Nick wants to merge 1 commit into
base: main
Choose a base branch
from
Open

fix(llm-pi-ai): 目录路由的模型发现改为合并线上清单 #309

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
203 changes: 201 additions & 2 deletions patches/@deepseek-ai+dsh-llm-pi-ai+0.1.2-rc.1.patch
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,8 +1,90 @@
diff --git a/node_modules/@deepseek-ai/dsh-llm-pi-ai/lib/index.js b/node_modules/@deepseek-ai/dsh-llm-pi-ai/lib/index.js
index 5c89b80..193f5b7 100644
index c25e5c7..8a52d95 100644
--- a/node_modules/@deepseek-ai/dsh-llm-pi-ai/lib/index.js
+++ b/node_modules/@deepseek-ai/dsh-llm-pi-ai/lib/index.js
@@ -1278,8 +1278,9 @@ function mapUsage(usage) {
@@ -511,17 +511,49 @@ function invalid(provider, detail) {
throw new Error(`llm-pi-ai: provider "${provider}" ${detail}`);
}
/**
-* The one wire protocol a catalog route's shipped models agree on. This is what
-* lets a deployment add a model the installed catalog has not caught up with —
-* a provider's newest release — without restating the protocol its siblings
-* already use. A route whose shipped models disagree (an OpenAI-style catalog
-* spanning Responses and Chat Completions) has no such answer, so a model it
-* does not describe must name its protocol at the route.
-*/
-function sharedCatalogApi(defaults) {
- const apis = /* @__PURE__ */ new Set();
- for (const model of defaults.values()) apis.add(model.api);
- return apis.size === 1 ? [...apis][0] : void 0;
+* The endpoint a catalog route's shipped models are dominated by: the wire
+* protocol the most of them speak, paired with the baseUrl those models agree
+* on. This is what lets a deployment add a model the installed catalog has not
+* caught up with — a provider's newest release — without restating endpoint
+* facts its siblings already carry.
+*
+* Dominance rather than unanimity, because a route whose shipped models
+* disagree (a subscription gateway spanning Chat Completions, Responses, and
+* Anthropic Messages) otherwise has no answer at all — and a `models` entry
+* cannot name its own protocol, so the only recourse left was a route-level
+* `api`, which wins over *every* sibling's and so breaks the ones speaking
+* another. Between guessing the plurality protocol for one new model and making
+* the route unable to name a new model without corrupting its siblings, the
+* plurality is the lesser error and the recoverable one: a model whose protocol
+* was guessed wrong is refused by its endpoint, named, on first use.
+*
+* @param defaults - the installed catalog models for the route.
+* @param listable - when given, only protocols in this set are considered.
+* @returns the dominant `api` and its `baseUrl`; empty when nothing qualifies.
+*/
+function catalogEndpoint(defaults, listable) {
+ const tally = /* @__PURE__ */ new Map();
+ for (const model of defaults.values()) {
+ if (listable !== void 0 && !listable.has(model.api)) continue;
+ const key = `${model.api}\n${model.baseUrl ?? ""}`;
+ tally.set(key, (tally.get(key) ?? 0) + 1);
+ }
+ let winner;
+ let winningCount = 0;
+ /* Insertion order is catalog order and `>` keeps the incumbent, so a tie
+ resolves to whichever pair the catalog lists first — stable across runs. */
+ for (const [key, count] of tally) if (count > winningCount) {
+ winner = key;
+ winningCount = count;
+ }
+ if (winner === void 0) return {};
+ const separator = winner.indexOf("\n");
+ const api = winner.slice(0, separator);
+ const baseUrl = winner.slice(separator + 1);
+ return {
+ api,
+ ...baseUrl.length === 0 ? {} : { baseUrl }
+ };
}
/**
* Resolve one model's reasoning capability from its declared efforts.
@@ -628,7 +660,7 @@ function resolveRouteModels(request) {
...overrides[model.id]
}));
if (entries.length === 0) invalid(provider, "resolves no models; the installed catalog does not describe this route, so its models must be listed in configuration");
- const routeApi = sharedCatalogApi(defaults);
+ const routeEndpoint = catalogEndpoint(defaults);
assertOfferedCompatFields(provider, "route", request.compat);
for (const entry of entries) assertOfferedCompatFields(provider, `model "${entry.id}"`, entry.compat);
const seen = /* @__PURE__ */ new Set();
@@ -638,9 +670,9 @@ function resolveRouteModels(request) {
if (seen.has(entry.id)) invalid(provider, `lists model "${entry.id}" more than once`);
seen.add(entry.id);
const base = defaults.get(entry.id);
- const api = request.api ?? base?.api ?? routeApi;
+ const api = request.api ?? base?.api ?? routeEndpoint.api;
if (api === void 0) invalid(provider, `model "${entry.id}" needs an api; the installed catalog does not describe it, so set the route's api to the wire protocol its endpoint speaks`);
- const baseUrl = request.baseURL ?? base?.baseUrl ?? providerBaseUrl;
+ const baseUrl = request.baseURL ?? base?.baseUrl ?? routeEndpoint.baseUrl ?? providerBaseUrl;
if (baseUrl === void 0) invalid(provider, `model "${entry.id}" needs a baseURL; the installed catalog does not describe this route`);
const contextWindow = entry.contextWindow ?? base?.contextWindow ?? request.defaultContextWindow;
if (!Number.isInteger(contextWindow) || contextWindow <= 0) invalid(provider, `model "${entry.id}" contextWindow must be a positive integer`);
@@ -1289,8 +1321,9 @@ function mapUsage(usage) {
};
}
function classifyPiAiError(message) {
Expand All @@ -13,3 +95,120 @@ index 5c89b80..193f5b7 100644
if (/\b429\b|rate.?limit/i.test(message)) return "RATE_LIMIT";
if (/\b413\b|failed to buffer the request body:\s*length limit exceeded|payload too large|request body too large/i.test(message)) return "INVALID_REQUEST";
if (/\b400\b|invalid.?request/i.test(message)) return "INVALID_REQUEST";
@@ -2017,11 +2050,20 @@ function authContextFrom(ctx) {
* Answering "which models can this provider serve?" for the configuration
* surface's "fetch available models" action.
*
-* A route the installed pi-ai catalog ships is answered **from that catalog**,
-* with no network call at all: pi-ai's registry is the authoritative list for
-* its own providers, and it carries the capacities a listing endpoint would
-* not disclose. Only a route the catalog does not describe — a gateway, a
-* self-hosted server — is interrogated over the wire.
+* A route the installed pi-ai catalog ships is answered from that catalog **and
+* from its endpoint**, the two merged. The catalog carries the capacities a
+* listing endpoint does not disclose, so it wins wherever both describe a
+* model; but the catalog is a build-time snapshot and a subscription gateway
+* adds models between pi-ai releases. Answering from the snapshot alone left
+* this action unable to ever report a model newer than the installed pi-ai —
+* the listing the user pressed the button for. A route the catalog does not
+* describe — a gateway, a self-hosted server — is answered from its endpoint
+* alone, as before.
+*
+* A catalog route whose endpoint cannot be read still answers from the
+* snapshot: an unreachable network, a refused key, or a protocol with no
+* readable listing leaves the shipped catalog as the better answer, and this
+* action has always been able to answer a catalog route offline.
*
* Neither path is a catalog refresh. Nothing here is stored: the request
* carries a draft the user is still editing, and the reply is candidate
@@ -2157,19 +2199,52 @@ function usableProbeKey(raw) {
* refuses or fails the request, or the reply is not a model listing.
*/
async function discoverModels(request, storedProfile) {
- if (request.provider !== void 0) {
- const installed = catalogModels(request.provider);
- if (installed.size > 0) return [...installed.values()].map((model) => ({
- id: model.id,
- name: model.name,
- contextWindow: model.contextWindow,
- maxTokens: model.maxTokens
- }));
+ const installed = request.provider === void 0 ? /* @__PURE__ */ new Map() : catalogModels(request.provider);
+ const shipped = () => [...installed.values()].map((model) => ({
+ id: model.id,
+ name: model.name,
+ contextWindow: model.contextWindow,
+ maxTokens: model.maxTokens
+ }));
+ if (installed.size === 0) return listEndpointModels(request, storedProfile, request.baseURL, request.api);
+ /* A catalog route's own endpoint facts, so the listing reaches the gateway
+ the route actually serves without the draft having to restate them. The
+ draft still wins: it is what the user is editing. */
+ const endpoint = catalogEndpoint(installed, LISTABLE_PROTOCOLS);
+ const baseURL = request.baseURL ?? endpoint.baseUrl;
+ const api = request.api ?? endpoint.api;
+ if (baseURL === void 0 || baseURL.length === 0 || api === void 0) return shipped();
+ let advertised;
+ try {
+ advertised = await listEndpointModels(request, storedProfile, baseURL, api);
+ } catch (error) {
+ /* Cancellation is the caller's own decision and must not be answered
+ with a stale list; every other failure — unreachable, refused, or
+ unreadable — leaves the shipped catalog as the better answer. */
+ if (error instanceof LlmError && error.code === "ABORTED") throw error;
+ return shipped();
}
- if (request.baseURL === void 0 || request.baseURL.length === 0) throw new LlmError(`pi-ai ships no catalog for provider "${request.provider ?? ""}", so its models can only come from its endpoint; set a baseURL, or enter this provider's models by hand`, "DISCOVERY_FAILED");
- const api = request.api ?? "openai-completions";
+ /* Catalog first: it carries capacities the listing omits. Endpoint order
+ decides where a model the catalog has not caught up with appears. */
+ const merged = new Map(advertised.map((model) => [model.id, model]));
+ for (const model of shipped()) merged.set(model.id, model);
+ return [...merged.values()];
+}
+/**
+* Read one endpoint's OpenAI-compatible model listing.
+* @param request - the draft being edited, for its credential and cancellation.
+* @param storedProfile - host-owned headers and lazy credential resolution.
+* @param baseURL - the endpoint to interrogate.
+* @param api - the wire protocol it speaks.
+* @returns the advertised models in endpoint order.
+* @throws LlmError when the protocol has no readable listing, the endpoint
+* refuses or fails the request, or the reply is not a model listing.
+*/
+async function listEndpointModels(request, storedProfile, baseURL, api$1) {
+ if (baseURL === void 0 || baseURL.length === 0) throw new LlmError(`pi-ai ships no catalog for provider "${request.provider ?? ""}", so its models can only come from its endpoint; set a baseURL, or enter this provider's models by hand`, "DISCOVERY_FAILED");
+ const api = api$1 ?? "openai-completions";
if (!LISTABLE_PROTOCOLS.has(api)) throw new LlmError(`pi-ai protocol "${api}" has no model listing this build can read; enter this provider's models by hand`, "DISCOVERY_UNSUPPORTED");
- const url = listingUrl(request.baseURL);
+ const url = listingUrl(baseURL);
const stored = storedProfile?.();
const supplied = request.apiKey ?? await stored?.resolveApiKey();
const apiKey = supplied === void 0 ? void 0 : usableProbeKey(supplied);
diff --git a/node_modules/@deepseek-ai/dsh-llm-pi-ai/lib/types/discovery.d.ts b/node_modules/@deepseek-ai/dsh-llm-pi-ai/lib/types/discovery.d.ts
index 97c5944..a6d4130 100644
--- a/node_modules/@deepseek-ai/dsh-llm-pi-ai/lib/types/discovery.d.ts
+++ b/node_modules/@deepseek-ai/dsh-llm-pi-ai/lib/types/discovery.d.ts
@@ -2,11 +2,17 @@
* Answering "which models can this provider serve?" for the configuration
* surface's "fetch available models" action.
*
- * A route the installed pi-ai catalog ships is answered **from that catalog**,
- * with no network call at all: pi-ai's registry is the authoritative list for
- * its own providers, and it carries the capacities a listing endpoint would
- * not disclose. Only a route the catalog does not describe — a gateway, a
- * self-hosted server — is interrogated over the wire.
+ * A route the installed pi-ai catalog ships is answered from that catalog **and
+ * from its endpoint**, the two merged. The catalog carries the capacities a
+ * listing endpoint does not disclose, so it wins wherever both describe a
+ * model; but the catalog is a build-time snapshot and a subscription gateway
+ * adds models between pi-ai releases, so the endpoint is what reports a model
+ * newer than the installed pi-ai. A route the catalog does not describe — a
+ * gateway, a self-hosted server — is answered from its endpoint alone.
+ *
+ * A catalog route whose endpoint cannot be read still answers from the
+ * snapshot: an unreachable network, a refused key, or a protocol with no
+ * readable listing leaves the shipped catalog as the better answer.
*
* Neither path is a catalog refresh. Nothing here is stored: the request
* carries a draft the user is still editing, and the reply is candidate
Loading