@huggingface/inference 4.0.3 → 4.0.5

package/README.md

@@ -651,9 +651,10 @@ You can use any Chat Completion API-compatible provider with the `chatCompletion
 ```typescript
 // Chat Completion Example
 const MISTRAL_KEY = process.env.MISTRAL_KEY;
-const hf = new InferenceClient(MISTRAL_KEY);
-const ep = hf.endpoint("https://api.mistral.ai");
-const stream = ep.chatCompletionStream({
+const hf = new InferenceClient(MISTRAL_KEY, {
+  endpointUrl: "https://api.mistral.ai",
+});
+const stream = hf.chatCompletionStream({
   model: "mistral-tiny",
   messages: [{ role: "user", content: "Complete the equation one + one = , just the answer" }],
 });

package/dist/commonjs/lib/getInferenceProviderMapping.d.ts

@@ -1,18 +1,19 @@
 import type { WidgetType } from "@huggingface/tasks";
 import type { InferenceProvider, InferenceProviderOrPolicy, ModelId } from "../types.js";
-export declare const inferenceProviderMappingCache: Map<string, Partial<Record<"black-forest-labs" | "cerebras" | "cohere" | "fal-ai" | "featherless-ai" | "fireworks-ai" | "groq" | "hf-inference" | "hyperbolic" | "nebius" | "novita" | "nscale" | "openai" | "ovhcloud" | "replicate" | "sambanova" | "together", Omit<InferenceProviderModelMapping, "hfModelId">>>>;
-export type InferenceProviderMapping = Partial<Record<InferenceProvider, Omit<InferenceProviderModelMapping, "hfModelId">>>;
-export interface InferenceProviderModelMapping {
+export declare const inferenceProviderMappingCache: Map<string, InferenceProviderMappingEntry[]>;
+export interface InferenceProviderMappingEntry {
     adapter?: string;
     adapterWeightsPath?: string;
     hfModelId: ModelId;
+    provider: string;
     providerId: string;
     status: "live" | "staging";
     task: WidgetType;
+    type?: "single-model" | "tag-filter";
 }
 export declare function fetchInferenceProviderMappingForModel(modelId: ModelId, accessToken?: string, options?: {
     fetch?: (input: RequestInfo, init?: RequestInit) => Promise<Response>;
-}): Promise<InferenceProviderMapping>;
+}): Promise<InferenceProviderMappingEntry[]>;
 export declare function getInferenceProviderMapping(params: {
     accessToken?: string;
     modelId: ModelId;
@@ -20,6 +21,6 @@ export declare function getInferenceProviderMapping(params: {
     task: WidgetType;
 }, options: {
     fetch?: (input: RequestInfo, init?: RequestInit) => Promise<Response>;
-}): Promise<InferenceProviderModelMapping | null>;
+}): Promise<InferenceProviderMappingEntry | null>;
 export declare function resolveProvider(provider?: InferenceProviderOrPolicy, modelId?: string, endpointUrl?: string): Promise<InferenceProvider>;
 //# sourceMappingURL=getInferenceProviderMapping.d.ts.map

package/dist/commonjs/lib/getInferenceProviderMapping.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"getInferenceProviderMapping.d.ts","sourceRoot":"","sources":["../../../src/lib/getInferenceProviderMapping.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,oBAAoB,CAAC;AAIrD,OAAO,KAAK,EAAE,iBAAiB,EAAE,yBAAyB,EAAE,OAAO,EAAE,MAAM,aAAa,CAAC;AAIzF,eAAO,MAAM,6BAA6B,uTAA+C,CAAC;AAE1F,MAAM,MAAM,wBAAwB,GAAG,OAAO,CAC7C,MAAM,CAAC,iBAAiB,EAAE,IAAI,CAAC,6BAA6B,EAAE,WAAW,CAAC,CAAC,CAC3E,CAAC;AAEF,MAAM,WAAW,6BAA6B;IAC7C,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,SAAS,EAAE,OAAO,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;IACnB,MAAM,EAAE,MAAM,GAAG,SAAS,CAAC;IAC3B,IAAI,EAAE,UAAU,CAAC;CACjB;AAED,wBAAsB,qCAAqC,CAC1D,OAAO,EAAE,OAAO,EAChB,WAAW,CAAC,EAAE,MAAM,EACpB,OAAO,CAAC,EAAE;IACT,KAAK,CAAC,EAAE,CAAC,KAAK,EAAE,WAAW,EAAE,IAAI,CAAC,EAAE,WAAW,KAAK,OAAO,CAAC,QAAQ,CAAC,CAAC;CACtE,GACC,OAAO,CAAC,wBAAwB,CAAC,CAgDnC;AAED,wBAAsB,2BAA2B,CAChD,MAAM,EAAE;IACP,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,OAAO,EAAE,OAAO,CAAC;IACjB,QAAQ,EAAE,iBAAiB,CAAC;IAC5B,IAAI,EAAE,UAAU,CAAC;CACjB,EACD,OAAO,EAAE;IACR,KAAK,CAAC,EAAE,CAAC,KAAK,EAAE,WAAW,EAAE,IAAI,CAAC,EAAE,WAAW,KAAK,OAAO,CAAC,QAAQ,CAAC,CAAC;CACtE,GACC,OAAO,CAAC,6BAA6B,GAAG,IAAI,CAAC,CA4B/C;AAED,wBAAsB,eAAe,CACpC,QAAQ,CAAC,EAAE,yBAAyB,EACpC,OAAO,CAAC,EAAE,MAAM,EAChB,WAAW,CAAC,EAAE,MAAM,GAClB,OAAO,CAAC,iBAAiB,CAAC,CAyB5B"}
+{"version":3,"file":"getInferenceProviderMapping.d.ts","sourceRoot":"","sources":["../../../src/lib/getInferenceProviderMapping.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,oBAAoB,CAAC;AAIrD,OAAO,KAAK,EAAE,iBAAiB,EAAE,yBAAyB,EAAE,OAAO,EAAE,MAAM,aAAa,CAAC;AAIzF,eAAO,MAAM,6BAA6B,8CAAsD,CAAC;AAEjG,MAAM,WAAW,6BAA6B;IAC7C,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,SAAS,EAAE,OAAO,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,UAAU,EAAE,MAAM,CAAC;IACnB,MAAM,EAAE,MAAM,GAAG,SAAS,CAAC;IAC3B,IAAI,EAAE,UAAU,CAAC;IACjB,IAAI,CAAC,EAAE,cAAc,GAAG,YAAY,CAAC;CACrC;AAiCD,wBAAsB,qCAAqC,CAC1D,OAAO,EAAE,OAAO,EAChB,WAAW,CAAC,EAAE,MAAM,EACpB,OAAO,CAAC,EAAE;IACT,KAAK,CAAC,EAAE,CAAC,KAAK,EAAE,WAAW,EAAE,IAAI,CAAC,EAAE,WAAW,KAAK,OAAO,CAAC,QAAQ,CAAC,CAAC;CACtE,GACC,OAAO,CAAC,6BAA6B,EAAE,CAAC,CAqD1C;AAED,wBAAsB,2BAA2B,CAChD,MAAM,EAAE;IACP,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,OAAO,EAAE,OAAO,CAAC;IACjB,QAAQ,EAAE,iBAAiB,CAAC;IAC5B,IAAI,EAAE,UAAU,CAAC;CACjB,EACD,OAAO,EAAE;IACR,KAAK,CAAC,EAAE,CAAC,KAAK,EAAE,WAAW,EAAE,IAAI,CAAC,EAAE,WAAW,KAAK,OAAO,CAAC,QAAQ,CAAC,CAAC;CACtE,GACC,OAAO,CAAC,6BAA6B,GAAG,IAAI,CAAC,CAwB/C;AAED,wBAAsB,eAAe,CACpC,QAAQ,CAAC,EAAE,yBAAyB,EACpC,OAAO,CAAC,EAAE,MAAM,EAChB,WAAW,CAAC,EAAE,MAAM,GAClB,OAAO,CAAC,iBAAiB,CAAC,CAyB5B"}

package/dist/commonjs/lib/getInferenceProviderMapping.js

@@ -10,6 +10,29 @@ const hf_inference_js_1 = require("../providers/hf-inference.js");
 const typedInclude_js_1 = require("../utils/typedInclude.js");
 const errors_js_1 = require("../errors.js");
 exports.inferenceProviderMappingCache = new Map();
+/**
+ * Normalize inferenceProviderMapping to always return an array format.
+ * This provides backward and forward compatibility for the API changes.
+ *
+ * Vendored from @huggingface/hub to avoid extra dependency.
+ */
+function normalizeInferenceProviderMapping(modelId, inferenceProviderMapping) {
+    if (!inferenceProviderMapping) {
+        return [];
+    }
+    // If it's already an array, return it as is
+    if (Array.isArray(inferenceProviderMapping)) {
+        return inferenceProviderMapping;
+    }
+    // Convert mapping to array format
+    return Object.entries(inferenceProviderMapping).map(([provider, mapping]) => ({
+        provider,
+        hfModelId: modelId,
+        providerId: mapping.providerId,
+        status: mapping.status,
+        task: mapping.task,
+    }));
+}
 async function fetchInferenceProviderMappingForModel(modelId, accessToken, options) {
     let inferenceProviderMapping;
     if (exports.inferenceProviderMappingCache.has(modelId)) {
@@ -42,7 +65,8 @@ async function fetchInferenceProviderMappingForModel(modelId, accessToken, optio
         if (!payload?.inferenceProviderMapping) {
             throw new errors_js_1.InferenceClientHubApiError(`We have not been able to find inference provider information for model ${modelId}.`, { url, method: "GET" }, { requestId: resp.headers.get("x-request-id") ?? "", status: resp.status, body: await resp.text() });
         }
-        inferenceProviderMapping = payload.inferenceProviderMapping;
+        inferenceProviderMapping = normalizeInferenceProviderMapping(modelId, payload.inferenceProviderMapping);
+        exports.inferenceProviderMappingCache.set(modelId, inferenceProviderMapping);
     }
     return inferenceProviderMapping;
 }
@@ -50,8 +74,8 @@ async function getInferenceProviderMapping(params, options) {
     if (consts_js_1.HARDCODED_MODEL_INFERENCE_MAPPING[params.provider][params.modelId]) {
         return consts_js_1.HARDCODED_MODEL_INFERENCE_MAPPING[params.provider][params.modelId];
     }
-    const inferenceProviderMapping = await fetchInferenceProviderMappingForModel(params.modelId, params.accessToken, options);
-    const providerMapping = inferenceProviderMapping[params.provider];
+    const mappings = await fetchInferenceProviderMappingForModel(params.modelId, params.accessToken, options);
+    const providerMapping = mappings.find((mapping) => mapping.provider === params.provider);
     if (providerMapping) {
         const equivalentTasks = params.provider === "hf-inference" && (0, typedInclude_js_1.typedInclude)(hf_inference_js_1.EQUIVALENT_SENTENCE_TRANSFORMERS_TASKS, params.task)
             ? hf_inference_js_1.EQUIVALENT_SENTENCE_TRANSFORMERS_TASKS
@@ -62,7 +86,7 @@ async function getInferenceProviderMapping(params, options) {
         if (providerMapping.status === "staging") {
             console.warn(`Model ${params.modelId} is in staging mode for provider ${params.provider}. Meant for test purposes only.`);
         }
-        return { ...providerMapping, hfModelId: params.modelId };
+        return providerMapping;
     }
     return null;
 }
@@ -82,8 +106,8 @@ async function resolveProvider(provider, modelId, endpointUrl) {
         if (!modelId) {
             throw new errors_js_1.InferenceClientInputError("Specifying a model is required when provider is 'auto'");
         }
-        const inferenceProviderMapping = await fetchInferenceProviderMappingForModel(modelId);
-        provider = Object.keys(inferenceProviderMapping)[0];
+        const mappings = await fetchInferenceProviderMappingForModel(modelId);
+        provider = mappings[0]?.provider;
     }
     if (!provider) {
         throw new errors_js_1.InferenceClientInputError(`No Inference Provider available for model ${modelId}.`);

package/dist/commonjs/lib/makeRequestOptions.d.ts

@@ -1,5 +1,5 @@
 import type { InferenceTask, Options, RequestArgs } from "../types.js";
-import type { InferenceProviderModelMapping } from "./getInferenceProviderMapping.js";
+import type { InferenceProviderMappingEntry } from "./getInferenceProviderMapping.js";
 import type { getProviderHelper } from "./getProviderHelper.js";
 /**
  * Helper that prepares request arguments.
@@ -22,7 +22,7 @@ export declare function makeRequestOptions(args: RequestArgs & {
 export declare function makeRequestOptionsFromResolvedModel(resolvedModel: string, providerHelper: ReturnType<typeof getProviderHelper>, args: RequestArgs & {
     data?: Blob | ArrayBuffer;
     stream?: boolean;
-}, mapping: InferenceProviderModelMapping | undefined, options?: Options & {
+}, mapping: InferenceProviderMappingEntry | undefined, options?: Options & {
     task?: InferenceTask;
 }): {
     url: string;

package/dist/commonjs/lib/makeRequestOptions.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"makeRequestOptions.d.ts","sourceRoot":"","sources":["../../../src/lib/makeRequestOptions.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,aAAa,EAAE,OAAO,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AACvE,OAAO,KAAK,EAAE,6BAA6B,EAAE,MAAM,kCAAkC,CAAC;AAEtF,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,wBAAwB,CAAC;AAUhE;;;GAGG;AACH,wBAAsB,kBAAkB,CACvC,IAAI,EAAE,WAAW,GAAG;IACnB,IAAI,CAAC,EAAE,IAAI,GAAG,WAAW,CAAC;IAC1B,MAAM,CAAC,EAAE,OAAO,CAAC;CACjB,EACD,cAAc,EAAE,UAAU,CAAC,OAAO,iBAAiB,CAAC,EACpD,OAAO,CAAC,EAAE,OAAO,GAAG;IACnB,oEAAoE;IACpE,IAAI,CAAC,EAAE,aAAa,CAAC;CACrB,GACC,OAAO,CAAC;IAAE,GAAG,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,WAAW,CAAA;CAAE,CAAC,CAqE7C;AAED;;;GAGG;AACH,wBAAgB,mCAAmC,CAClD,aAAa,EAAE,MAAM,EACrB,cAAc,EAAE,UAAU,CAAC,OAAO,iBAAiB,CAAC,EACpD,IAAI,EAAE,WAAW,GAAG;IACnB,IAAI,CAAC,EAAE,IAAI,GAAG,WAAW,CAAC;IAC1B,MAAM,CAAC,EAAE,OAAO,CAAC;CACjB,EACD,OAAO,EAAE,6BAA6B,GAAG,SAAS,EAClD,OAAO,CAAC,EAAE,OAAO,GAAG;IACnB,IAAI,CAAC,EAAE,aAAa,CAAC;CACrB,GACC;IAAE,GAAG,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,WAAW,CAAA;CAAE,CA8EpC"}
+{"version":3,"file":"makeRequestOptions.d.ts","sourceRoot":"","sources":["../../../src/lib/makeRequestOptions.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,aAAa,EAAE,OAAO,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AACvE,OAAO,KAAK,EAAE,6BAA6B,EAAE,MAAM,kCAAkC,CAAC;AAEtF,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,wBAAwB,CAAC;AAUhE;;;GAGG;AACH,wBAAsB,kBAAkB,CACvC,IAAI,EAAE,WAAW,GAAG;IACnB,IAAI,CAAC,EAAE,IAAI,GAAG,WAAW,CAAC;IAC1B,MAAM,CAAC,EAAE,OAAO,CAAC;CACjB,EACD,cAAc,EAAE,UAAU,CAAC,OAAO,iBAAiB,CAAC,EACpD,OAAO,CAAC,EAAE,OAAO,GAAG;IACnB,oEAAoE;IACpE,IAAI,CAAC,EAAE,aAAa,CAAC;CACrB,GACC,OAAO,CAAC;IAAE,GAAG,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,WAAW,CAAA;CAAE,CAAC,CAsE7C;AAED;;;GAGG;AACH,wBAAgB,mCAAmC,CAClD,aAAa,EAAE,MAAM,EACrB,cAAc,EAAE,UAAU,CAAC,OAAO,iBAAiB,CAAC,EACpD,IAAI,EAAE,WAAW,GAAG;IACnB,IAAI,CAAC,EAAE,IAAI,GAAG,WAAW,CAAC;IAC1B,MAAM,CAAC,EAAE,OAAO,CAAC;CACjB,EACD,OAAO,EAAE,6BAA6B,GAAG,SAAS,EAClD,OAAO,CAAC,EAAE,OAAO,GAAG;IACnB,IAAI,CAAC,EAAE,aAAa,CAAC;CACrB,GACC;IAAE,GAAG,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,WAAW,CAAA;CAAE,CA8EpC"}

package/dist/commonjs/lib/makeRequestOptions.js

@@ -41,6 +41,7 @@ async function makeRequestOptions(args, providerHelper, options) {
     }
     const inferenceProviderMapping = providerHelper.clientSideRoutingOnly
         ? {
+            provider: provider,
             // eslint-disable-next-line @typescript-eslint/no-non-null-assertion
             providerId: removeProviderPrefix(maybeModel, provider),
             // eslint-disable-next-line @typescript-eslint/no-non-null-assertion

package/dist/commonjs/package.d.ts

@@ -1,3 +1,3 @@
-export declare const PACKAGE_VERSION = "4.0.3";
+export declare const PACKAGE_VERSION = "4.0.5";
 export declare const PACKAGE_NAME = "@huggingface/inference";
 //# sourceMappingURL=package.d.ts.map

package/dist/commonjs/package.js

@@ -2,5 +2,5 @@
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.PACKAGE_NAME = exports.PACKAGE_VERSION = void 0;
 // Generated file from package.json. Issues importing JSON directly when publishing on commonjs/ESM - see https://github.com/microsoft/TypeScript/issues/51783
-exports.PACKAGE_VERSION = "4.0.3";
+exports.PACKAGE_VERSION = "4.0.5";
 exports.PACKAGE_NAME = "@huggingface/inference";

package/dist/commonjs/providers/consts.d.ts

@@ -1,4 +1,4 @@
-import type { InferenceProviderModelMapping } from "../lib/getInferenceProviderMapping.js";
+import type { InferenceProviderMappingEntry } from "../lib/getInferenceProviderMapping.js";
 import type { InferenceProvider } from "../types.js";
 import { type ModelId } from "../types.js";
 /**
@@ -8,5 +8,5 @@ import { type ModelId } from "../types.js";
  *
  * We also inject into this dictionary from tests.
  */
-export declare const HARDCODED_MODEL_INFERENCE_MAPPING: Record<InferenceProvider, Record<ModelId, InferenceProviderModelMapping>>;
+export declare const HARDCODED_MODEL_INFERENCE_MAPPING: Record<InferenceProvider, Record<ModelId, InferenceProviderMappingEntry>>;
 //# sourceMappingURL=consts.d.ts.map

package/dist/commonjs/snippets/getInferenceSnippets.d.ts

@@ -1,11 +1,12 @@
 import { type InferenceSnippet, type ModelDataMinimal } from "@huggingface/tasks";
-import type { InferenceProviderModelMapping } from "../lib/getInferenceProviderMapping.js";
+import type { InferenceProviderMappingEntry } from "../lib/getInferenceProviderMapping.js";
 import type { InferenceProviderOrPolicy } from "../types.js";
 export type InferenceSnippetOptions = {
     streaming?: boolean;
     billTo?: string;
     accessToken?: string;
     directRequest?: boolean;
+    endpointUrl?: string;
 } & Record<string, unknown>;
-export declare function getInferenceSnippets(model: ModelDataMinimal, provider: InferenceProviderOrPolicy, inferenceProviderMapping?: InferenceProviderModelMapping, opts?: Record<string, unknown>): InferenceSnippet[];
+export declare function getInferenceSnippets(model: ModelDataMinimal, provider: InferenceProviderOrPolicy, inferenceProviderMapping?: InferenceProviderMappingEntry, opts?: Record<string, unknown>): InferenceSnippet[];
 //# sourceMappingURL=getInferenceSnippets.d.ts.map

package/dist/commonjs/snippets/getInferenceSnippets.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"getInferenceSnippets.d.ts","sourceRoot":"","sources":["../../../src/snippets/getInferenceSnippets.ts"],"names":[],"mappings":"AACA,OAAO,EACN,KAAK,gBAAgB,EAErB,KAAK,gBAAgB,EAGrB,MAAM,oBAAoB,CAAC;AAG5B,OAAO,KAAK,EAAE,6BAA6B,EAAE,MAAM,uCAAuC,CAAC;AAG3F,OAAO,KAAK,EAAE,yBAAyB,EAA8B,MAAM,aAAa,CAAC;AAGzF,MAAM,MAAM,uBAAuB,GAAG;IACrC,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,aAAa,CAAC,EAAE,OAAO,CAAC;CACxB,GAAG,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;AAiV5B,wBAAgB,oBAAoB,CACnC,KAAK,EAAE,gBAAgB,EACvB,QAAQ,EAAE,yBAAyB,EACnC,wBAAwB,CAAC,EAAE,6BAA6B,EACxD,IAAI,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAC5B,gBAAgB,EAAE,CAIpB"}
+{"version":3,"file":"getInferenceSnippets.d.ts","sourceRoot":"","sources":["../../../src/snippets/getInferenceSnippets.ts"],"names":[],"mappings":"AACA,OAAO,EACN,KAAK,gBAAgB,EAErB,KAAK,gBAAgB,EAGrB,MAAM,oBAAoB,CAAC;AAG5B,OAAO,KAAK,EAAE,6BAA6B,EAAE,MAAM,uCAAuC,CAAC;AAG3F,OAAO,KAAK,EAAE,yBAAyB,EAA8B,MAAM,aAAa,CAAC;AAGzF,MAAM,MAAM,uBAAuB,GAAG;IACrC,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,WAAW,CAAC,EAAE,MAAM,CAAC;CACrB,GAAG,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;AA2V5B,wBAAgB,oBAAoB,CACnC,KAAK,EAAE,gBAAgB,EACvB,QAAQ,EAAE,yBAAyB,EACnC,wBAAwB,CAAC,EAAE,6BAA6B,EACxD,IAAI,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAC5B,gBAAgB,EAAE,CAIpB"}

package/dist/commonjs/snippets/getInferenceSnippets.js

@@ -110,6 +110,7 @@ const snippetGenerator = (templateName, inputPreparationFn) => {
         const request = (0, makeRequestOptions_js_1.makeRequestOptionsFromResolvedModel)(providerModelId, providerHelper, {
             accessToken: accessTokenOrPlaceholder,
             provider,
+            endpointUrl: opts?.endpointUrl,
             ...inputs,
         }, inferenceProviderMapping, {
             task,
@@ -151,6 +152,7 @@ const snippetGenerator = (templateName, inputPreparationFn) => {
             provider,
             providerModelId: providerModelId ?? model.id,
             billTo: opts?.billTo,
+            endpointUrl: opts?.endpointUrl,
         };
         /// Iterate over clients => check if a snippet exists => generate
         const clients = provider === "auto" ? CLIENTS_AUTO_POLICY : CLIENTS;
@@ -195,7 +197,7 @@ const snippetGenerator = (templateName, inputPreparationFn) => {
                 }
                 /// Replace access token placeholder
                 if (snippet.includes(placeholder)) {
-                    snippet = replaceAccessTokenPlaceholder(opts?.directRequest, placeholder, snippet, language, provider);
+                    snippet = replaceAccessTokenPlaceholder(opts?.directRequest, placeholder, snippet, language, provider, opts?.endpointUrl);
                 }
                 /// Snippet is ready!
                 return { language, client: client, content: snippet };
@@ -322,17 +324,20 @@ function indentString(str) {
 function removeSuffix(str, suffix) {
     return str.endsWith(suffix) ? str.slice(0, -suffix.length) : str;
 }
-function replaceAccessTokenPlaceholder(directRequest, placeholder, snippet, language, provider) {
+function replaceAccessTokenPlaceholder(directRequest, placeholder, snippet, language, provider, endpointUrl) {
     // If "opts.accessToken" is not set, the snippets are generated with a placeholder.
     // Once snippets are rendered, we replace the placeholder with code to fetch the access token from an environment variable.
     // Determine if HF_TOKEN or specific provider token should be used
-    const useHfToken = provider == "hf-inference" || // hf-inference provider => use $HF_TOKEN
-        (!directRequest && // if explicit directRequest => use provider-specific token
-            (!snippet.includes("https://") || // no URL provided => using a client => use $HF_TOKEN
-                snippet.includes("https://router.huggingface.co"))); // explicit routed request => use $HF_TOKEN
+    const useHfToken = !endpointUrl && // custom endpointUrl => use a generic API_TOKEN
+        (provider == "hf-inference" || // hf-inference provider => use $HF_TOKEN
+            (!directRequest && // if explicit directRequest => use provider-specific token
+                (snippet.includes("InferenceClient") || // using a client => use $HF_TOKEN
+                    snippet.includes("https://router.huggingface.co")))); // explicit routed request => use $HF_TOKEN
     const accessTokenEnvVar = useHfToken
         ? "HF_TOKEN" // e.g. routed request or hf-inference
-        : provider.toUpperCase().replace("-", "_") + "_API_KEY"; // e.g. "REPLICATE_API_KEY"
+        : endpointUrl
+            ? "API_TOKEN"
+            : provider.toUpperCase().replace("-", "_") + "_API_KEY"; // e.g. "REPLICATE_API_KEY"
     // Replace the placeholder with the env variable
     if (language === "sh") {
         snippet = snippet.replace(`'Authorization: Bearer ${placeholder}'`, `"Authorization: Bearer $${accessTokenEnvVar}"` // e.g. "Authorization: Bearer $HF_TOKEN"

package/dist/commonjs/snippets/templates.exported.js

@@ -14,14 +14,14 @@ exports.templates = {
             "zeroShotClassification": "async function query(data) {\n    const response = await fetch(\n\t\t\"{{ fullUrl }}\",\n        {\n            headers: {\n\t\t\t\tAuthorization: \"{{ authorizationHeader }}\",\n                \"Content-Type\": \"application/json\",\n{% if billTo %}\n                \"X-HF-Bill-To\": \"{{ billTo }}\",\n{% endif %}         },\n            method: \"POST\",\n            body: JSON.stringify(data),\n        }\n    );\n    const result = await response.json();\n    return result;\n}\n\nquery({\n    inputs: {{ providerInputs.asObj.inputs }},\n    parameters: { candidate_labels: [\"refund\", \"legal\", \"faq\"] }\n}).then((response) => {\n    console.log(JSON.stringify(response));\n});"
         },
         "huggingface.js": {
-            "basic": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst output = await client.{{ methodName }}({\n\tmodel: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n\tprovider: \"{{ provider }}\",\n}{% if billTo %}, {\n\tbillTo: \"{{ billTo }}\",\n}{% endif %});\n\nconsole.log(output);",
-            "basicAudio": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst data = fs.readFileSync({{inputs.asObj.inputs}});\n\nconst output = await client.{{ methodName }}({\n\tdata,\n\tmodel: \"{{ model.id }}\",\n\tprovider: \"{{ provider }}\",\n}{% if billTo %}, {\n\tbillTo: \"{{ billTo }}\",\n}{% endif %});\n\nconsole.log(output);",
-            "basicImage": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst data = fs.readFileSync({{inputs.asObj.inputs}});\n\nconst output = await client.{{ methodName }}({\n\tdata,\n\tmodel: \"{{ model.id }}\",\n\tprovider: \"{{ provider }}\",\n}{% if billTo %}, {\n\tbillTo: \"{{ billTo }}\",\n}{% endif %});\n\nconsole.log(output);",
-            "conversational": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst chatCompletion = await client.chatCompletion({\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n{{ inputs.asTsString }}\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n\nconsole.log(chatCompletion.choices[0].message);",
-            "conversationalStream": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nlet out = \"\";\n\nconst stream = client.chatCompletionStream({\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n{{ inputs.asTsString }}\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n\nfor await (const chunk of stream) {\n\tif (chunk.choices && chunk.choices.length > 0) {\n\t\tconst newContent = chunk.choices[0].delta.content;\n\t\tout += newContent;\n\t\tconsole.log(newContent);\n\t}\n}",
-            "textToImage": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst image = await client.textToImage({\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n\tparameters: { num_inference_steps: 5 },\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n/// Use the generated image (it's a Blob)",
-            "textToSpeech": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst audio = await client.textToSpeech({\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n// Use the generated audio (it's a Blob)",
-            "textToVideo": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst video = await client.textToVideo({\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n// Use the generated video (it's a Blob)"
+            "basic": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst output = await client.{{ methodName }}({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n\tmodel: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n\tprovider: \"{{ provider }}\",\n}{% if billTo %}, {\n\tbillTo: \"{{ billTo }}\",\n}{% endif %});\n\nconsole.log(output);",
+            "basicAudio": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst data = fs.readFileSync({{inputs.asObj.inputs}});\n\nconst output = await client.{{ methodName }}({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n\tdata,\n\tmodel: \"{{ model.id }}\",\n\tprovider: \"{{ provider }}\",\n}{% if billTo %}, {\n\tbillTo: \"{{ billTo }}\",\n}{% endif %});\n\nconsole.log(output);",
+            "basicImage": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst data = fs.readFileSync({{inputs.asObj.inputs}});\n\nconst output = await client.{{ methodName }}({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n\tdata,\n\tmodel: \"{{ model.id }}\",\n\tprovider: \"{{ provider }}\",\n}{% if billTo %}, {\n\tbillTo: \"{{ billTo }}\",\n}{% endif %});\n\nconsole.log(output);",
+            "conversational": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst chatCompletion = await client.chatCompletion({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n{{ inputs.asTsString }}\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n\nconsole.log(chatCompletion.choices[0].message);",
+            "conversationalStream": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nlet out = \"\";\n\nconst stream = client.chatCompletionStream({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n{{ inputs.asTsString }}\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n\nfor await (const chunk of stream) {\n\tif (chunk.choices && chunk.choices.length > 0) {\n\t\tconst newContent = chunk.choices[0].delta.content;\n\t\tout += newContent;\n\t\tconsole.log(newContent);\n\t}\n}",
+            "textToImage": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst image = await client.textToImage({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n\tparameters: { num_inference_steps: 5 },\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n/// Use the generated image (it's a Blob)",
+            "textToSpeech": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst audio = await client.textToSpeech({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n// Use the generated audio (it's a Blob)",
+            "textToVideo": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst video = await client.textToVideo({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n// Use the generated video (it's a Blob)"
         },
         "openai": {
             "conversational": "import { OpenAI } from \"openai\";\n\nconst client = new OpenAI({\n\tbaseURL: \"{{ baseUrl }}\",\n\tapiKey: \"{{ accessToken }}\",\n{% if billTo %}\n\tdefaultHeaders: {\n\t\t\"X-HF-Bill-To\": \"{{ billTo }}\" \n\t}\n{% endif %}\n});\n\nconst chatCompletion = await client.chat.completions.create({\n\tmodel: \"{{ providerModelId }}\",\n{{ inputs.asTsString }}\n});\n\nconsole.log(chatCompletion.choices[0].message);",
@@ -40,7 +40,7 @@ exports.templates = {
             "conversationalStream": "stream = client.chat.completions.create(\n    model=\"{{ model.id }}\",\n{{ inputs.asPythonString }}\n    stream=True,\n)\n\nfor chunk in stream:\n    print(chunk.choices[0].delta.content, end=\"\") ",
             "documentQuestionAnswering": "output = client.document_question_answering(\n    \"{{ inputs.asObj.image }}\",\n    question=\"{{ inputs.asObj.question }}\",\n    model=\"{{ model.id }}\",\n) ",
             "imageToImage": "# output is a PIL.Image object\nimage = client.image_to_image(\n    \"{{ inputs.asObj.inputs }}\",\n    prompt=\"{{ inputs.asObj.parameters.prompt }}\",\n    model=\"{{ model.id }}\",\n) ",
-            "importInferenceClient": "from huggingface_hub import InferenceClient\n\nclient = InferenceClient(\n    provider=\"{{ provider }}\",\n    api_key=\"{{ accessToken }}\",\n{% if billTo %}\n    bill_to=\"{{ billTo }}\",\n{% endif %}\n)",
+            "importInferenceClient": "from huggingface_hub import InferenceClient\n\nclient = InferenceClient(\n{% if endpointUrl %}\n    base_url=\"{{ baseUrl }}\",\n{% endif %}\n    provider=\"{{ provider }}\",\n    api_key=\"{{ accessToken }}\",\n{% if billTo %}\n    bill_to=\"{{ billTo }}\",\n{% endif %}\n)",
             "questionAnswering": "answer = client.question_answering(\n    question=\"{{ inputs.asObj.question }}\",\n    context=\"{{ inputs.asObj.context }}\",\n    model=\"{{ model.id }}\",\n) ",
             "tableQuestionAnswering": "answer = client.table_question_answering(\n    query=\"{{ inputs.asObj.query }}\",\n    table={{ inputs.asObj.table }},\n    model=\"{{ model.id }}\",\n) ",
             "textToImage": "# output is a PIL.Image object\nimage = client.text_to_image(\n    {{ inputs.asObj.inputs }},\n    model=\"{{ model.id }}\",\n) ",

package/dist/commonjs/types.d.ts

@@ -1,5 +1,5 @@
 import type { ChatCompletionInput, PipelineType } from "@huggingface/tasks";
-import type { InferenceProviderModelMapping } from "./lib/getInferenceProviderMapping.js";
+import type { InferenceProviderMappingEntry } from "./lib/getInferenceProviderMapping.js";
 /**
  * HF model id, like "meta-llama/Llama-3.3-70B-Instruct"
  */
@@ -91,7 +91,7 @@ export interface UrlParams {
 export interface BodyParams<T extends Record<string, unknown> = Record<string, unknown>> {
     args: T;
     model: string;
-    mapping?: InferenceProviderModelMapping | undefined;
+    mapping?: InferenceProviderMappingEntry | undefined;
     task?: InferenceTask;
 }
 //# sourceMappingURL=types.d.ts.map

package/dist/esm/lib/getInferenceProviderMapping.d.ts

@@ -1,18 +1,19 @@
 import type { WidgetType } from "@huggingface/tasks";
 import type { InferenceProvider, InferenceProviderOrPolicy, ModelId } from "../types.js";
-export declare const inferenceProviderMappingCache: Map<string, Partial<Record<"black-forest-labs" | "cerebras" | "cohere" | "fal-ai" | "featherless-ai" | "fireworks-ai" | "groq" | "hf-inference" | "hyperbolic" | "nebius" | "novita" | "nscale" | "openai" | "ovhcloud" | "replicate" | "sambanova" | "together", Omit<InferenceProviderModelMapping, "hfModelId">>>>;
-export type InferenceProviderMapping = Partial<Record<InferenceProvider, Omit<InferenceProviderModelMapping, "hfModelId">>>;
-export interface InferenceProviderModelMapping {
+export declare const inferenceProviderMappingCache: Map<string, InferenceProviderMappingEntry[]>;
+export interface InferenceProviderMappingEntry {
     adapter?: string;
     adapterWeightsPath?: string;
     hfModelId: ModelId;
+    provider: string;
     providerId: string;
     status: "live" | "staging";
     task: WidgetType;
+    type?: "single-model" | "tag-filter";
 }
 export declare function fetchInferenceProviderMappingForModel(modelId: ModelId, accessToken?: string, options?: {
     fetch?: (input: RequestInfo, init?: RequestInit) => Promise<Response>;
-}): Promise<InferenceProviderMapping>;
+}): Promise<InferenceProviderMappingEntry[]>;
 export declare function getInferenceProviderMapping(params: {
     accessToken?: string;
     modelId: ModelId;
@@ -20,6 +21,6 @@ export declare function getInferenceProviderMapping(params: {
     task: WidgetType;
 }, options: {
     fetch?: (input: RequestInfo, init?: RequestInit) => Promise<Response>;
-}): Promise<InferenceProviderModelMapping | null>;
+}): Promise<InferenceProviderMappingEntry | null>;
 export declare function resolveProvider(provider?: InferenceProviderOrPolicy, modelId?: string, endpointUrl?: string): Promise<InferenceProvider>;
 //# sourceMappingURL=getInferenceProviderMapping.d.ts.map

package/dist/esm/lib/getInferenceProviderMapping.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"getInferenceProviderMapping.d.ts","sourceRoot":"","sources":["../../../src/lib/getInferenceProviderMapping.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,oBAAoB,CAAC;AAIrD,OAAO,KAAK,EAAE,iBAAiB,EAAE,yBAAyB,EAAE,OAAO,EAAE,MAAM,aAAa,CAAC;AAIzF,eAAO,MAAM,6BAA6B,uTAA+C,CAAC;AAE1F,MAAM,MAAM,wBAAwB,GAAG,OAAO,CAC7C,MAAM,CAAC,iBAAiB,EAAE,IAAI,CAAC,6BAA6B,EAAE,WAAW,CAAC,CAAC,CAC3E,CAAC;AAEF,MAAM,WAAW,6BAA6B;IAC7C,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,SAAS,EAAE,OAAO,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;IACnB,MAAM,EAAE,MAAM,GAAG,SAAS,CAAC;IAC3B,IAAI,EAAE,UAAU,CAAC;CACjB;AAED,wBAAsB,qCAAqC,CAC1D,OAAO,EAAE,OAAO,EAChB,WAAW,CAAC,EAAE,MAAM,EACpB,OAAO,CAAC,EAAE;IACT,KAAK,CAAC,EAAE,CAAC,KAAK,EAAE,WAAW,EAAE,IAAI,CAAC,EAAE,WAAW,KAAK,OAAO,CAAC,QAAQ,CAAC,CAAC;CACtE,GACC,OAAO,CAAC,wBAAwB,CAAC,CAgDnC;AAED,wBAAsB,2BAA2B,CAChD,MAAM,EAAE;IACP,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,OAAO,EAAE,OAAO,CAAC;IACjB,QAAQ,EAAE,iBAAiB,CAAC;IAC5B,IAAI,EAAE,UAAU,CAAC;CACjB,EACD,OAAO,EAAE;IACR,KAAK,CAAC,EAAE,CAAC,KAAK,EAAE,WAAW,EAAE,IAAI,CAAC,EAAE,WAAW,KAAK,OAAO,CAAC,QAAQ,CAAC,CAAC;CACtE,GACC,OAAO,CAAC,6BAA6B,GAAG,IAAI,CAAC,CA4B/C;AAED,wBAAsB,eAAe,CACpC,QAAQ,CAAC,EAAE,yBAAyB,EACpC,OAAO,CAAC,EAAE,MAAM,EAChB,WAAW,CAAC,EAAE,MAAM,GAClB,OAAO,CAAC,iBAAiB,CAAC,CAyB5B"}
+{"version":3,"file":"getInferenceProviderMapping.d.ts","sourceRoot":"","sources":["../../../src/lib/getInferenceProviderMapping.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,oBAAoB,CAAC;AAIrD,OAAO,KAAK,EAAE,iBAAiB,EAAE,yBAAyB,EAAE,OAAO,EAAE,MAAM,aAAa,CAAC;AAIzF,eAAO,MAAM,6BAA6B,8CAAsD,CAAC;AAEjG,MAAM,WAAW,6BAA6B;IAC7C,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,SAAS,EAAE,OAAO,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,UAAU,EAAE,MAAM,CAAC;IACnB,MAAM,EAAE,MAAM,GAAG,SAAS,CAAC;IAC3B,IAAI,EAAE,UAAU,CAAC;IACjB,IAAI,CAAC,EAAE,cAAc,GAAG,YAAY,CAAC;CACrC;AAiCD,wBAAsB,qCAAqC,CAC1D,OAAO,EAAE,OAAO,EAChB,WAAW,CAAC,EAAE,MAAM,EACpB,OAAO,CAAC,EAAE;IACT,KAAK,CAAC,EAAE,CAAC,KAAK,EAAE,WAAW,EAAE,IAAI,CAAC,EAAE,WAAW,KAAK,OAAO,CAAC,QAAQ,CAAC,CAAC;CACtE,GACC,OAAO,CAAC,6BAA6B,EAAE,CAAC,CAqD1C;AAED,wBAAsB,2BAA2B,CAChD,MAAM,EAAE;IACP,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,OAAO,EAAE,OAAO,CAAC;IACjB,QAAQ,EAAE,iBAAiB,CAAC;IAC5B,IAAI,EAAE,UAAU,CAAC;CACjB,EACD,OAAO,EAAE;IACR,KAAK,CAAC,EAAE,CAAC,KAAK,EAAE,WAAW,EAAE,IAAI,CAAC,EAAE,WAAW,KAAK,OAAO,CAAC,QAAQ,CAAC,CAAC;CACtE,GACC,OAAO,CAAC,6BAA6B,GAAG,IAAI,CAAC,CAwB/C;AAED,wBAAsB,eAAe,CACpC,QAAQ,CAAC,EAAE,yBAAyB,EACpC,OAAO,CAAC,EAAE,MAAM,EAChB,WAAW,CAAC,EAAE,MAAM,GAClB,OAAO,CAAC,iBAAiB,CAAC,CAyB5B"}

package/dist/esm/lib/getInferenceProviderMapping.js

@@ -4,6 +4,29 @@ import { EQUIVALENT_SENTENCE_TRANSFORMERS_TASKS } from "../providers/hf-inferenc
 import { typedInclude } from "../utils/typedInclude.js";
 import { InferenceClientHubApiError, InferenceClientInputError } from "../errors.js";
 export const inferenceProviderMappingCache = new Map();
+/**
+ * Normalize inferenceProviderMapping to always return an array format.
+ * This provides backward and forward compatibility for the API changes.
+ *
+ * Vendored from @huggingface/hub to avoid extra dependency.
+ */
+function normalizeInferenceProviderMapping(modelId, inferenceProviderMapping) {
+    if (!inferenceProviderMapping) {
+        return [];
+    }
+    // If it's already an array, return it as is
+    if (Array.isArray(inferenceProviderMapping)) {
+        return inferenceProviderMapping;
+    }
+    // Convert mapping to array format
+    return Object.entries(inferenceProviderMapping).map(([provider, mapping]) => ({
+        provider,
+        hfModelId: modelId,
+        providerId: mapping.providerId,
+        status: mapping.status,
+        task: mapping.task,
+    }));
+}
 export async function fetchInferenceProviderMappingForModel(modelId, accessToken, options) {
     let inferenceProviderMapping;
     if (inferenceProviderMappingCache.has(modelId)) {
@@ -36,7 +59,8 @@ export async function fetchInferenceProviderMappingForModel(modelId, accessToken
         if (!payload?.inferenceProviderMapping) {
             throw new InferenceClientHubApiError(`We have not been able to find inference provider information for model ${modelId}.`, { url, method: "GET" }, { requestId: resp.headers.get("x-request-id") ?? "", status: resp.status, body: await resp.text() });
         }
-        inferenceProviderMapping = payload.inferenceProviderMapping;
+        inferenceProviderMapping = normalizeInferenceProviderMapping(modelId, payload.inferenceProviderMapping);
+        inferenceProviderMappingCache.set(modelId, inferenceProviderMapping);
     }
     return inferenceProviderMapping;
 }
@@ -44,8 +68,8 @@ export async function getInferenceProviderMapping(params, options) {
     if (HARDCODED_MODEL_INFERENCE_MAPPING[params.provider][params.modelId]) {
         return HARDCODED_MODEL_INFERENCE_MAPPING[params.provider][params.modelId];
     }
-    const inferenceProviderMapping = await fetchInferenceProviderMappingForModel(params.modelId, params.accessToken, options);
-    const providerMapping = inferenceProviderMapping[params.provider];
+    const mappings = await fetchInferenceProviderMappingForModel(params.modelId, params.accessToken, options);
+    const providerMapping = mappings.find((mapping) => mapping.provider === params.provider);
     if (providerMapping) {
         const equivalentTasks = params.provider === "hf-inference" && typedInclude(EQUIVALENT_SENTENCE_TRANSFORMERS_TASKS, params.task)
             ? EQUIVALENT_SENTENCE_TRANSFORMERS_TASKS
@@ -56,7 +80,7 @@ export async function getInferenceProviderMapping(params, options) {
         if (providerMapping.status === "staging") {
             console.warn(`Model ${params.modelId} is in staging mode for provider ${params.provider}. Meant for test purposes only.`);
         }
-        return { ...providerMapping, hfModelId: params.modelId };
+        return providerMapping;
     }
     return null;
 }
@@ -76,8 +100,8 @@ export async function resolveProvider(provider, modelId, endpointUrl) {
         if (!modelId) {
             throw new InferenceClientInputError("Specifying a model is required when provider is 'auto'");
         }
-        const inferenceProviderMapping = await fetchInferenceProviderMappingForModel(modelId);
-        provider = Object.keys(inferenceProviderMapping)[0];
+        const mappings = await fetchInferenceProviderMappingForModel(modelId);
+        provider = mappings[0]?.provider;
     }
     if (!provider) {
         throw new InferenceClientInputError(`No Inference Provider available for model ${modelId}.`);

package/dist/esm/lib/makeRequestOptions.d.ts

@@ -1,5 +1,5 @@
 import type { InferenceTask, Options, RequestArgs } from "../types.js";
-import type { InferenceProviderModelMapping } from "./getInferenceProviderMapping.js";
+import type { InferenceProviderMappingEntry } from "./getInferenceProviderMapping.js";
 import type { getProviderHelper } from "./getProviderHelper.js";
 /**
  * Helper that prepares request arguments.
@@ -22,7 +22,7 @@ export declare function makeRequestOptions(args: RequestArgs & {
 export declare function makeRequestOptionsFromResolvedModel(resolvedModel: string, providerHelper: ReturnType<typeof getProviderHelper>, args: RequestArgs & {
     data?: Blob | ArrayBuffer;
     stream?: boolean;
-}, mapping: InferenceProviderModelMapping | undefined, options?: Options & {
+}, mapping: InferenceProviderMappingEntry | undefined, options?: Options & {
     task?: InferenceTask;
 }): {
     url: string;

package/dist/esm/lib/makeRequestOptions.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"makeRequestOptions.d.ts","sourceRoot":"","sources":["../../../src/lib/makeRequestOptions.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,aAAa,EAAE,OAAO,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AACvE,OAAO,KAAK,EAAE,6BAA6B,EAAE,MAAM,kCAAkC,CAAC;AAEtF,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,wBAAwB,CAAC;AAUhE;;;GAGG;AACH,wBAAsB,kBAAkB,CACvC,IAAI,EAAE,WAAW,GAAG;IACnB,IAAI,CAAC,EAAE,IAAI,GAAG,WAAW,CAAC;IAC1B,MAAM,CAAC,EAAE,OAAO,CAAC;CACjB,EACD,cAAc,EAAE,UAAU,CAAC,OAAO,iBAAiB,CAAC,EACpD,OAAO,CAAC,EAAE,OAAO,GAAG;IACnB,oEAAoE;IACpE,IAAI,CAAC,EAAE,aAAa,CAAC;CACrB,GACC,OAAO,CAAC;IAAE,GAAG,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,WAAW,CAAA;CAAE,CAAC,CAqE7C;AAED;;;GAGG;AACH,wBAAgB,mCAAmC,CAClD,aAAa,EAAE,MAAM,EACrB,cAAc,EAAE,UAAU,CAAC,OAAO,iBAAiB,CAAC,EACpD,IAAI,EAAE,WAAW,GAAG;IACnB,IAAI,CAAC,EAAE,IAAI,GAAG,WAAW,CAAC;IAC1B,MAAM,CAAC,EAAE,OAAO,CAAC;CACjB,EACD,OAAO,EAAE,6BAA6B,GAAG,SAAS,EAClD,OAAO,CAAC,EAAE,OAAO,GAAG;IACnB,IAAI,CAAC,EAAE,aAAa,CAAC;CACrB,GACC;IAAE,GAAG,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,WAAW,CAAA;CAAE,CA8EpC"}
+{"version":3,"file":"makeRequestOptions.d.ts","sourceRoot":"","sources":["../../../src/lib/makeRequestOptions.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,aAAa,EAAE,OAAO,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AACvE,OAAO,KAAK,EAAE,6BAA6B,EAAE,MAAM,kCAAkC,CAAC;AAEtF,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,wBAAwB,CAAC;AAUhE;;;GAGG;AACH,wBAAsB,kBAAkB,CACvC,IAAI,EAAE,WAAW,GAAG;IACnB,IAAI,CAAC,EAAE,IAAI,GAAG,WAAW,CAAC;IAC1B,MAAM,CAAC,EAAE,OAAO,CAAC;CACjB,EACD,cAAc,EAAE,UAAU,CAAC,OAAO,iBAAiB,CAAC,EACpD,OAAO,CAAC,EAAE,OAAO,GAAG;IACnB,oEAAoE;IACpE,IAAI,CAAC,EAAE,aAAa,CAAC;CACrB,GACC,OAAO,CAAC;IAAE,GAAG,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,WAAW,CAAA;CAAE,CAAC,CAsE7C;AAED;;;GAGG;AACH,wBAAgB,mCAAmC,CAClD,aAAa,EAAE,MAAM,EACrB,cAAc,EAAE,UAAU,CAAC,OAAO,iBAAiB,CAAC,EACpD,IAAI,EAAE,WAAW,GAAG;IACnB,IAAI,CAAC,EAAE,IAAI,GAAG,WAAW,CAAC;IAC1B,MAAM,CAAC,EAAE,OAAO,CAAC;CACjB,EACD,OAAO,EAAE,6BAA6B,GAAG,SAAS,EAClD,OAAO,CAAC,EAAE,OAAO,GAAG;IACnB,IAAI,CAAC,EAAE,aAAa,CAAC;CACrB,GACC;IAAE,GAAG,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,WAAW,CAAA;CAAE,CA8EpC"}

package/dist/esm/lib/makeRequestOptions.js

@@ -37,6 +37,7 @@ export async function makeRequestOptions(args, providerHelper, options) {
     }
     const inferenceProviderMapping = providerHelper.clientSideRoutingOnly
         ? {
+            provider: provider,
             // eslint-disable-next-line @typescript-eslint/no-non-null-assertion
             providerId: removeProviderPrefix(maybeModel, provider),
             // eslint-disable-next-line @typescript-eslint/no-non-null-assertion

package/dist/esm/package.d.ts

@@ -1,3 +1,3 @@
-export declare const PACKAGE_VERSION = "4.0.3";
+export declare const PACKAGE_VERSION = "4.0.5";
 export declare const PACKAGE_NAME = "@huggingface/inference";
 //# sourceMappingURL=package.d.ts.map

package/dist/esm/package.js

@@ -1,3 +1,3 @@
 // Generated file from package.json. Issues importing JSON directly when publishing on commonjs/ESM - see https://github.com/microsoft/TypeScript/issues/51783
-export const PACKAGE_VERSION = "4.0.3";
+export const PACKAGE_VERSION = "4.0.5";
 export const PACKAGE_NAME = "@huggingface/inference";

package/dist/esm/providers/consts.d.ts

@@ -1,4 +1,4 @@
-import type { InferenceProviderModelMapping } from "../lib/getInferenceProviderMapping.js";
+import type { InferenceProviderMappingEntry } from "../lib/getInferenceProviderMapping.js";
 import type { InferenceProvider } from "../types.js";
 import { type ModelId } from "../types.js";
 /**
@@ -8,5 +8,5 @@ import { type ModelId } from "../types.js";
  *
  * We also inject into this dictionary from tests.
  */
-export declare const HARDCODED_MODEL_INFERENCE_MAPPING: Record<InferenceProvider, Record<ModelId, InferenceProviderModelMapping>>;
+export declare const HARDCODED_MODEL_INFERENCE_MAPPING: Record<InferenceProvider, Record<ModelId, InferenceProviderMappingEntry>>;
 //# sourceMappingURL=consts.d.ts.map

package/dist/esm/snippets/getInferenceSnippets.d.ts

@@ -1,11 +1,12 @@
 import { type InferenceSnippet, type ModelDataMinimal } from "@huggingface/tasks";
-import type { InferenceProviderModelMapping } from "../lib/getInferenceProviderMapping.js";
+import type { InferenceProviderMappingEntry } from "../lib/getInferenceProviderMapping.js";
 import type { InferenceProviderOrPolicy } from "../types.js";
 export type InferenceSnippetOptions = {
     streaming?: boolean;
     billTo?: string;
     accessToken?: string;
     directRequest?: boolean;
+    endpointUrl?: string;
 } & Record<string, unknown>;
-export declare function getInferenceSnippets(model: ModelDataMinimal, provider: InferenceProviderOrPolicy, inferenceProviderMapping?: InferenceProviderModelMapping, opts?: Record<string, unknown>): InferenceSnippet[];
+export declare function getInferenceSnippets(model: ModelDataMinimal, provider: InferenceProviderOrPolicy, inferenceProviderMapping?: InferenceProviderMappingEntry, opts?: Record<string, unknown>): InferenceSnippet[];
 //# sourceMappingURL=getInferenceSnippets.d.ts.map

package/dist/esm/snippets/getInferenceSnippets.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"getInferenceSnippets.d.ts","sourceRoot":"","sources":["../../../src/snippets/getInferenceSnippets.ts"],"names":[],"mappings":"AACA,OAAO,EACN,KAAK,gBAAgB,EAErB,KAAK,gBAAgB,EAGrB,MAAM,oBAAoB,CAAC;AAG5B,OAAO,KAAK,EAAE,6BAA6B,EAAE,MAAM,uCAAuC,CAAC;AAG3F,OAAO,KAAK,EAAE,yBAAyB,EAA8B,MAAM,aAAa,CAAC;AAGzF,MAAM,MAAM,uBAAuB,GAAG;IACrC,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,aAAa,CAAC,EAAE,OAAO,CAAC;CACxB,GAAG,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;AAiV5B,wBAAgB,oBAAoB,CACnC,KAAK,EAAE,gBAAgB,EACvB,QAAQ,EAAE,yBAAyB,EACnC,wBAAwB,CAAC,EAAE,6BAA6B,EACxD,IAAI,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAC5B,gBAAgB,EAAE,CAIpB"}
+{"version":3,"file":"getInferenceSnippets.d.ts","sourceRoot":"","sources":["../../../src/snippets/getInferenceSnippets.ts"],"names":[],"mappings":"AACA,OAAO,EACN,KAAK,gBAAgB,EAErB,KAAK,gBAAgB,EAGrB,MAAM,oBAAoB,CAAC;AAG5B,OAAO,KAAK,EAAE,6BAA6B,EAAE,MAAM,uCAAuC,CAAC;AAG3F,OAAO,KAAK,EAAE,yBAAyB,EAA8B,MAAM,aAAa,CAAC;AAGzF,MAAM,MAAM,uBAAuB,GAAG;IACrC,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,WAAW,CAAC,EAAE,MAAM,CAAC;CACrB,GAAG,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;AA2V5B,wBAAgB,oBAAoB,CACnC,KAAK,EAAE,gBAAgB,EACvB,QAAQ,EAAE,yBAAyB,EACnC,wBAAwB,CAAC,EAAE,6BAA6B,EACxD,IAAI,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAC5B,gBAAgB,EAAE,CAIpB"}

package/dist/esm/snippets/getInferenceSnippets.js

@@ -107,6 +107,7 @@ const snippetGenerator = (templateName, inputPreparationFn) => {
         const request = makeRequestOptionsFromResolvedModel(providerModelId, providerHelper, {
             accessToken: accessTokenOrPlaceholder,
             provider,
+            endpointUrl: opts?.endpointUrl,
             ...inputs,
         }, inferenceProviderMapping, {
             task,
@@ -148,6 +149,7 @@ const snippetGenerator = (templateName, inputPreparationFn) => {
             provider,
             providerModelId: providerModelId ?? model.id,
             billTo: opts?.billTo,
+            endpointUrl: opts?.endpointUrl,
         };
         /// Iterate over clients => check if a snippet exists => generate
         const clients = provider === "auto" ? CLIENTS_AUTO_POLICY : CLIENTS;
@@ -192,7 +194,7 @@ const snippetGenerator = (templateName, inputPreparationFn) => {
                 }
                 /// Replace access token placeholder
                 if (snippet.includes(placeholder)) {
-                    snippet = replaceAccessTokenPlaceholder(opts?.directRequest, placeholder, snippet, language, provider);
+                    snippet = replaceAccessTokenPlaceholder(opts?.directRequest, placeholder, snippet, language, provider, opts?.endpointUrl);
                 }
                 /// Snippet is ready!
                 return { language, client: client, content: snippet };
@@ -319,17 +321,20 @@ function indentString(str) {
 function removeSuffix(str, suffix) {
     return str.endsWith(suffix) ? str.slice(0, -suffix.length) : str;
 }
-function replaceAccessTokenPlaceholder(directRequest, placeholder, snippet, language, provider) {
+function replaceAccessTokenPlaceholder(directRequest, placeholder, snippet, language, provider, endpointUrl) {
     // If "opts.accessToken" is not set, the snippets are generated with a placeholder.
     // Once snippets are rendered, we replace the placeholder with code to fetch the access token from an environment variable.
     // Determine if HF_TOKEN or specific provider token should be used
-    const useHfToken = provider == "hf-inference" || // hf-inference provider => use $HF_TOKEN
-        (!directRequest && // if explicit directRequest => use provider-specific token
-            (!snippet.includes("https://") || // no URL provided => using a client => use $HF_TOKEN
-                snippet.includes("https://router.huggingface.co"))); // explicit routed request => use $HF_TOKEN
+    const useHfToken = !endpointUrl && // custom endpointUrl => use a generic API_TOKEN
+        (provider == "hf-inference" || // hf-inference provider => use $HF_TOKEN
+            (!directRequest && // if explicit directRequest => use provider-specific token
+                (snippet.includes("InferenceClient") || // using a client => use $HF_TOKEN
+                    snippet.includes("https://router.huggingface.co")))); // explicit routed request => use $HF_TOKEN
     const accessTokenEnvVar = useHfToken
         ? "HF_TOKEN" // e.g. routed request or hf-inference
-        : provider.toUpperCase().replace("-", "_") + "_API_KEY"; // e.g. "REPLICATE_API_KEY"
+        : endpointUrl
+            ? "API_TOKEN"
+            : provider.toUpperCase().replace("-", "_") + "_API_KEY"; // e.g. "REPLICATE_API_KEY"
     // Replace the placeholder with the env variable
     if (language === "sh") {
         snippet = snippet.replace(`'Authorization: Bearer ${placeholder}'`, `"Authorization: Bearer $${accessTokenEnvVar}"` // e.g. "Authorization: Bearer $HF_TOKEN"

package/dist/esm/snippets/templates.exported.js

@@ -11,14 +11,14 @@ export const templates = {
             "zeroShotClassification": "async function query(data) {\n    const response = await fetch(\n\t\t\"{{ fullUrl }}\",\n        {\n            headers: {\n\t\t\t\tAuthorization: \"{{ authorizationHeader }}\",\n                \"Content-Type\": \"application/json\",\n{% if billTo %}\n                \"X-HF-Bill-To\": \"{{ billTo }}\",\n{% endif %}         },\n            method: \"POST\",\n            body: JSON.stringify(data),\n        }\n    );\n    const result = await response.json();\n    return result;\n}\n\nquery({\n    inputs: {{ providerInputs.asObj.inputs }},\n    parameters: { candidate_labels: [\"refund\", \"legal\", \"faq\"] }\n}).then((response) => {\n    console.log(JSON.stringify(response));\n});"
         },
         "huggingface.js": {
-            "basic": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst output = await client.{{ methodName }}({\n\tmodel: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n\tprovider: \"{{ provider }}\",\n}{% if billTo %}, {\n\tbillTo: \"{{ billTo }}\",\n}{% endif %});\n\nconsole.log(output);",
-            "basicAudio": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst data = fs.readFileSync({{inputs.asObj.inputs}});\n\nconst output = await client.{{ methodName }}({\n\tdata,\n\tmodel: \"{{ model.id }}\",\n\tprovider: \"{{ provider }}\",\n}{% if billTo %}, {\n\tbillTo: \"{{ billTo }}\",\n}{% endif %});\n\nconsole.log(output);",
-            "basicImage": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst data = fs.readFileSync({{inputs.asObj.inputs}});\n\nconst output = await client.{{ methodName }}({\n\tdata,\n\tmodel: \"{{ model.id }}\",\n\tprovider: \"{{ provider }}\",\n}{% if billTo %}, {\n\tbillTo: \"{{ billTo }}\",\n}{% endif %});\n\nconsole.log(output);",
-            "conversational": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst chatCompletion = await client.chatCompletion({\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n{{ inputs.asTsString }}\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n\nconsole.log(chatCompletion.choices[0].message);",
-            "conversationalStream": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nlet out = \"\";\n\nconst stream = client.chatCompletionStream({\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n{{ inputs.asTsString }}\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n\nfor await (const chunk of stream) {\n\tif (chunk.choices && chunk.choices.length > 0) {\n\t\tconst newContent = chunk.choices[0].delta.content;\n\t\tout += newContent;\n\t\tconsole.log(newContent);\n\t}\n}",
-            "textToImage": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst image = await client.textToImage({\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n\tparameters: { num_inference_steps: 5 },\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n/// Use the generated image (it's a Blob)",
-            "textToSpeech": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst audio = await client.textToSpeech({\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n// Use the generated audio (it's a Blob)",
-            "textToVideo": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst video = await client.textToVideo({\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n// Use the generated video (it's a Blob)"
+            "basic": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst output = await client.{{ methodName }}({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n\tmodel: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n\tprovider: \"{{ provider }}\",\n}{% if billTo %}, {\n\tbillTo: \"{{ billTo }}\",\n}{% endif %});\n\nconsole.log(output);",
+            "basicAudio": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst data = fs.readFileSync({{inputs.asObj.inputs}});\n\nconst output = await client.{{ methodName }}({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n\tdata,\n\tmodel: \"{{ model.id }}\",\n\tprovider: \"{{ provider }}\",\n}{% if billTo %}, {\n\tbillTo: \"{{ billTo }}\",\n}{% endif %});\n\nconsole.log(output);",
+            "basicImage": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst data = fs.readFileSync({{inputs.asObj.inputs}});\n\nconst output = await client.{{ methodName }}({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n\tdata,\n\tmodel: \"{{ model.id }}\",\n\tprovider: \"{{ provider }}\",\n}{% if billTo %}, {\n\tbillTo: \"{{ billTo }}\",\n}{% endif %});\n\nconsole.log(output);",
+            "conversational": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst chatCompletion = await client.chatCompletion({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n{{ inputs.asTsString }}\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n\nconsole.log(chatCompletion.choices[0].message);",
+            "conversationalStream": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nlet out = \"\";\n\nconst stream = client.chatCompletionStream({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n{{ inputs.asTsString }}\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n\nfor await (const chunk of stream) {\n\tif (chunk.choices && chunk.choices.length > 0) {\n\t\tconst newContent = chunk.choices[0].delta.content;\n\t\tout += newContent;\n\t\tconsole.log(newContent);\n\t}\n}",
+            "textToImage": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst image = await client.textToImage({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n\tparameters: { num_inference_steps: 5 },\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n/// Use the generated image (it's a Blob)",
+            "textToSpeech": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst audio = await client.textToSpeech({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n// Use the generated audio (it's a Blob)",
+            "textToVideo": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst video = await client.textToVideo({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n// Use the generated video (it's a Blob)"
         },
         "openai": {
             "conversational": "import { OpenAI } from \"openai\";\n\nconst client = new OpenAI({\n\tbaseURL: \"{{ baseUrl }}\",\n\tapiKey: \"{{ accessToken }}\",\n{% if billTo %}\n\tdefaultHeaders: {\n\t\t\"X-HF-Bill-To\": \"{{ billTo }}\" \n\t}\n{% endif %}\n});\n\nconst chatCompletion = await client.chat.completions.create({\n\tmodel: \"{{ providerModelId }}\",\n{{ inputs.asTsString }}\n});\n\nconsole.log(chatCompletion.choices[0].message);",
@@ -37,7 +37,7 @@ export const templates = {
             "conversationalStream": "stream = client.chat.completions.create(\n    model=\"{{ model.id }}\",\n{{ inputs.asPythonString }}\n    stream=True,\n)\n\nfor chunk in stream:\n    print(chunk.choices[0].delta.content, end=\"\") ",
             "documentQuestionAnswering": "output = client.document_question_answering(\n    \"{{ inputs.asObj.image }}\",\n    question=\"{{ inputs.asObj.question }}\",\n    model=\"{{ model.id }}\",\n) ",
             "imageToImage": "# output is a PIL.Image object\nimage = client.image_to_image(\n    \"{{ inputs.asObj.inputs }}\",\n    prompt=\"{{ inputs.asObj.parameters.prompt }}\",\n    model=\"{{ model.id }}\",\n) ",
-            "importInferenceClient": "from huggingface_hub import InferenceClient\n\nclient = InferenceClient(\n    provider=\"{{ provider }}\",\n    api_key=\"{{ accessToken }}\",\n{% if billTo %}\n    bill_to=\"{{ billTo }}\",\n{% endif %}\n)",
+            "importInferenceClient": "from huggingface_hub import InferenceClient\n\nclient = InferenceClient(\n{% if endpointUrl %}\n    base_url=\"{{ baseUrl }}\",\n{% endif %}\n    provider=\"{{ provider }}\",\n    api_key=\"{{ accessToken }}\",\n{% if billTo %}\n    bill_to=\"{{ billTo }}\",\n{% endif %}\n)",
             "questionAnswering": "answer = client.question_answering(\n    question=\"{{ inputs.asObj.question }}\",\n    context=\"{{ inputs.asObj.context }}\",\n    model=\"{{ model.id }}\",\n) ",
             "tableQuestionAnswering": "answer = client.table_question_answering(\n    query=\"{{ inputs.asObj.query }}\",\n    table={{ inputs.asObj.table }},\n    model=\"{{ model.id }}\",\n) ",
             "textToImage": "# output is a PIL.Image object\nimage = client.text_to_image(\n    {{ inputs.asObj.inputs }},\n    model=\"{{ model.id }}\",\n) ",

package/dist/esm/types.d.ts

@@ -1,5 +1,5 @@
 import type { ChatCompletionInput, PipelineType } from "@huggingface/tasks";
-import type { InferenceProviderModelMapping } from "./lib/getInferenceProviderMapping.js";
+import type { InferenceProviderMappingEntry } from "./lib/getInferenceProviderMapping.js";
 /**
  * HF model id, like "meta-llama/Llama-3.3-70B-Instruct"
  */
@@ -91,7 +91,7 @@ export interface UrlParams {
 export interface BodyParams<T extends Record<string, unknown> = Record<string, unknown>> {
     args: T;
     model: string;
-    mapping?: InferenceProviderModelMapping | undefined;
+    mapping?: InferenceProviderMappingEntry | undefined;
     task?: InferenceTask;
 }
 //# sourceMappingURL=types.d.ts.map

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@huggingface/inference",
-  "version": "4.0.3",
+  "version": "4.0.5",
   "license": "MIT",
   "author": "Hugging Face and Tim Mikeladze <tim.mikeladze@gmail.com>",
   "description": "Typescript client for the Hugging Face Inference Providers and Inference Endpoints",
@@ -40,7 +40,7 @@
   },
   "type": "module",
   "dependencies": {
-    "@huggingface/tasks": "^0.19.12",
+    "@huggingface/tasks": "^0.19.15",
     "@huggingface/jinja": "^0.5.0"
   },
   "devDependencies": {

package/src/lib/getInferenceProviderMapping.ts

@@ -6,19 +6,48 @@ import type { InferenceProvider, InferenceProviderOrPolicy, ModelId } from "../t
 import { typedInclude } from "../utils/typedInclude.js";
 import { InferenceClientHubApiError, InferenceClientInputError } from "../errors.js";
 
-export const inferenceProviderMappingCache = new Map<ModelId, InferenceProviderMapping>();
+export const inferenceProviderMappingCache = new Map<ModelId, InferenceProviderMappingEntry[]>();
 
-export type InferenceProviderMapping = Partial<
-	Record<InferenceProvider, Omit<InferenceProviderModelMapping, "hfModelId">>
->;
-
-export interface InferenceProviderModelMapping {
+export interface InferenceProviderMappingEntry {
 	adapter?: string;
 	adapterWeightsPath?: string;
 	hfModelId: ModelId;
+	provider: string;
 	providerId: string;
 	status: "live" | "staging";
 	task: WidgetType;
+	type?: "single-model" | "tag-filter";
+}
+
+/**
+ * Normalize inferenceProviderMapping to always return an array format.
+ * This provides backward and forward compatibility for the API changes.
+ *
+ * Vendored from @huggingface/hub to avoid extra dependency.
+ */
+function normalizeInferenceProviderMapping(
+	modelId: ModelId,
+	inferenceProviderMapping?:
+		| InferenceProviderMappingEntry[]
+		| Record<string, { providerId: string; status: "live" | "staging"; task: WidgetType }>
+): InferenceProviderMappingEntry[] {
+	if (!inferenceProviderMapping) {
+		return [];
+	}
+
+	// If it's already an array, return it as is
+	if (Array.isArray(inferenceProviderMapping)) {
+		return inferenceProviderMapping;
+	}
+
+	// Convert mapping to array format
+	return Object.entries(inferenceProviderMapping).map(([provider, mapping]) => ({
+		provider,
+		hfModelId: modelId,
+		providerId: mapping.providerId,
+		status: mapping.status,
+		task: mapping.task,
+	}));
 }
 
 export async function fetchInferenceProviderMappingForModel(
@@ -27,8 +56,8 @@ export async function fetchInferenceProviderMappingForModel(
 	options?: {
 		fetch?: (input: RequestInfo, init?: RequestInit) => Promise<Response>;
 	}
-): Promise<InferenceProviderMapping> {
-	let inferenceProviderMapping: InferenceProviderMapping | null;
+): Promise<InferenceProviderMappingEntry[]> {
+	let inferenceProviderMapping: InferenceProviderMappingEntry[] | null;
 	if (inferenceProviderMappingCache.has(modelId)) {
 		// eslint-disable-next-line @typescript-eslint/no-non-null-assertion
 		inferenceProviderMapping = inferenceProviderMappingCache.get(modelId)!;
@@ -55,7 +84,11 @@ export async function fetchInferenceProviderMappingForModel(
 				);
 			}
 		}
-		let payload: { inferenceProviderMapping?: InferenceProviderMapping } | null = null;
+		let payload: {
+			inferenceProviderMapping?:
+				| InferenceProviderMappingEntry[]
+				| Record<string, { providerId: string; status: "live" | "staging"; task: WidgetType }>;
+		} | null = null;
 		try {
 			payload = await resp.json();
 		} catch {
@@ -72,7 +105,8 @@ export async function fetchInferenceProviderMappingForModel(
 				{ requestId: resp.headers.get("x-request-id") ?? "", status: resp.status, body: await resp.text() }
 			);
 		}
-		inferenceProviderMapping = payload.inferenceProviderMapping;
+		inferenceProviderMapping = normalizeInferenceProviderMapping(modelId, payload.inferenceProviderMapping);
+		inferenceProviderMappingCache.set(modelId, inferenceProviderMapping);
 	}
 	return inferenceProviderMapping;
 }
@@ -87,16 +121,12 @@ export async function getInferenceProviderMapping(
 	options: {
 		fetch?: (input: RequestInfo, init?: RequestInit) => Promise<Response>;
 	}
-): Promise<InferenceProviderModelMapping | null> {
+): Promise<InferenceProviderMappingEntry | null> {
 	if (HARDCODED_MODEL_INFERENCE_MAPPING[params.provider][params.modelId]) {
 		return HARDCODED_MODEL_INFERENCE_MAPPING[params.provider][params.modelId];
 	}
-	const inferenceProviderMapping = await fetchInferenceProviderMappingForModel(
-		params.modelId,
-		params.accessToken,
-		options
-	);
-	const providerMapping = inferenceProviderMapping[params.provider];
+	const mappings = await fetchInferenceProviderMappingForModel(params.modelId, params.accessToken, options);
+	const providerMapping = mappings.find((mapping) => mapping.provider === params.provider);
 	if (providerMapping) {
 		const equivalentTasks =
 			params.provider === "hf-inference" && typedInclude(EQUIVALENT_SENTENCE_TRANSFORMERS_TASKS, params.task)
@@ -112,7 +142,7 @@ export async function getInferenceProviderMapping(
 				`Model ${params.modelId} is in staging mode for provider ${params.provider}. Meant for test purposes only.`
 			);
 		}
-		return { ...providerMapping, hfModelId: params.modelId };
+		return providerMapping;
 	}
 	return null;
 }
@@ -139,8 +169,8 @@ export async function resolveProvider(
 		if (!modelId) {
 			throw new InferenceClientInputError("Specifying a model is required when provider is 'auto'");
 		}
-		const inferenceProviderMapping = await fetchInferenceProviderMappingForModel(modelId);
-		provider = Object.keys(inferenceProviderMapping)[0] as InferenceProvider | undefined;
+		const mappings = await fetchInferenceProviderMappingForModel(modelId);
+		provider = mappings[0]?.provider as InferenceProvider | undefined;
 	}
 	if (!provider) {
 		throw new InferenceClientInputError(`No Inference Provider available for model ${modelId}.`);

package/src/lib/makeRequestOptions.ts

@@ -1,7 +1,7 @@
 import { HF_HEADER_X_BILL_TO, HF_HUB_URL } from "../config.js";
 import { PACKAGE_NAME, PACKAGE_VERSION } from "../package.js";
 import type { InferenceTask, Options, RequestArgs } from "../types.js";
-import type { InferenceProviderModelMapping } from "./getInferenceProviderMapping.js";
+import type { InferenceProviderMappingEntry } from "./getInferenceProviderMapping.js";
 import { getInferenceProviderMapping } from "./getInferenceProviderMapping.js";
 import type { getProviderHelper } from "./getProviderHelper.js";
 import { isUrl } from "./isUrl.js";
@@ -64,6 +64,7 @@ export async function makeRequestOptions(
 
 	const inferenceProviderMapping = providerHelper.clientSideRoutingOnly
 		? ({
+				provider: provider,
 				// eslint-disable-next-line @typescript-eslint/no-non-null-assertion
 				providerId: removeProviderPrefix(maybeModel!, provider),
 				// eslint-disable-next-line @typescript-eslint/no-non-null-assertion
@@ -71,7 +72,7 @@ export async function makeRequestOptions(
 				status: "live",
 				// eslint-disable-next-line @typescript-eslint/no-non-null-assertion
 				task: task!,
-		  } satisfies InferenceProviderModelMapping)
+		  } satisfies InferenceProviderMappingEntry)
 		: await getInferenceProviderMapping(
 				{
 					modelId: hfModel,
@@ -109,7 +110,7 @@ export function makeRequestOptionsFromResolvedModel(
 		data?: Blob | ArrayBuffer;
 		stream?: boolean;
 	},
-	mapping: InferenceProviderModelMapping | undefined,
+	mapping: InferenceProviderMappingEntry | undefined,
 	options?: Options & {
 		task?: InferenceTask;
 	}

package/src/package.ts

@@ -1,3 +1,3 @@
 // Generated file from package.json. Issues importing JSON directly when publishing on commonjs/ESM - see https://github.com/microsoft/TypeScript/issues/51783
-export const PACKAGE_VERSION = "4.0.3";
+export const PACKAGE_VERSION = "4.0.5";
 export const PACKAGE_NAME = "@huggingface/inference";

package/src/providers/consts.ts

@@ -1,4 +1,4 @@
-import type { InferenceProviderModelMapping } from "../lib/getInferenceProviderMapping.js";
+import type { InferenceProviderMappingEntry } from "../lib/getInferenceProviderMapping.js";
 import type { InferenceProvider } from "../types.js";
 import { type ModelId } from "../types.js";
 
@@ -11,7 +11,7 @@ import { type ModelId } from "../types.js";
  */
 export const HARDCODED_MODEL_INFERENCE_MAPPING: Record<
 	InferenceProvider,
-	Record<ModelId, InferenceProviderModelMapping>
+	Record<ModelId, InferenceProviderMappingEntry>
 > = {
 	/**
 	 * "HF model ID" => "Model ID on Inference Provider's side"

package/src/snippets/getInferenceSnippets.ts

@@ -8,7 +8,7 @@ import {
 } from "@huggingface/tasks";
 import type { PipelineType, WidgetType } from "@huggingface/tasks";
 import type { ChatCompletionInputMessage, GenerationParameters } from "@huggingface/tasks";
-import type { InferenceProviderModelMapping } from "../lib/getInferenceProviderMapping.js";
+import type { InferenceProviderMappingEntry } from "../lib/getInferenceProviderMapping.js";
 import { getProviderHelper } from "../lib/getProviderHelper.js";
 import { makeRequestOptionsFromResolvedModel } from "../lib/makeRequestOptions.js";
 import type { InferenceProviderOrPolicy, InferenceTask, RequestArgs } from "../types.js";
@@ -18,7 +18,8 @@ export type InferenceSnippetOptions = {
 	streaming?: boolean;
 	billTo?: string;
 	accessToken?: string;
-	directRequest?: boolean;
+	directRequest?: boolean; // to bypass HF routing and call the provider directly
+	endpointUrl?: string; // to call a local endpoint directly
 } & Record<string, unknown>;
 
 const PYTHON_CLIENTS = ["huggingface_hub", "fal_client", "requests", "openai"] as const;
@@ -53,6 +54,7 @@ interface TemplateParams {
 	methodName?: string; // specific to snippetBasic
 	importBase64?: boolean; // specific to snippetImportRequests
 	importJson?: boolean; // specific to snippetImportRequests
+	endpointUrl?: string;
 }
 
 // Helpers to find + load templates
@@ -136,7 +138,7 @@ const snippetGenerator = (templateName: string, inputPreparationFn?: InputPrepar
 	return (
 		model: ModelDataMinimal,
 		provider: InferenceProviderOrPolicy,
-		inferenceProviderMapping?: InferenceProviderModelMapping,
+		inferenceProviderMapping?: InferenceProviderMappingEntry,
 		opts?: InferenceSnippetOptions
 	): InferenceSnippet[] => {
 		const providerModelId = inferenceProviderMapping?.providerId ?? model.id;
@@ -172,6 +174,7 @@ const snippetGenerator = (templateName: string, inputPreparationFn?: InputPrepar
 			{
 				accessToken: accessTokenOrPlaceholder,
 				provider,
+				endpointUrl: opts?.endpointUrl,
 				...inputs,
 			} as RequestArgs,
 			inferenceProviderMapping,
@@ -217,6 +220,7 @@ const snippetGenerator = (templateName: string, inputPreparationFn?: InputPrepar
 			provider,
 			providerModelId: providerModelId ?? model.id,
 			billTo: opts?.billTo,
+			endpointUrl: opts?.endpointUrl,
 		};
 
 		/// Iterate over clients => check if a snippet exists => generate
@@ -265,7 +269,14 @@ const snippetGenerator = (templateName: string, inputPreparationFn?: InputPrepar
 
 						/// Replace access token placeholder
 						if (snippet.includes(placeholder)) {
-							snippet = replaceAccessTokenPlaceholder(opts?.directRequest, placeholder, snippet, language, provider);
+							snippet = replaceAccessTokenPlaceholder(
+								opts?.directRequest,
+								placeholder,
+								snippet,
+								language,
+								provider,
+								opts?.endpointUrl
+							);
 						}
 
 						/// Snippet is ready!
@@ -320,7 +331,7 @@ const snippets: Partial<
 		(
 			model: ModelDataMinimal,
 			provider: InferenceProviderOrPolicy,
-			inferenceProviderMapping?: InferenceProviderModelMapping,
+			inferenceProviderMapping?: InferenceProviderMappingEntry,
 			opts?: InferenceSnippetOptions
 		) => InferenceSnippet[]
 	>
@@ -359,7 +370,7 @@ const snippets: Partial<
 export function getInferenceSnippets(
 	model: ModelDataMinimal,
 	provider: InferenceProviderOrPolicy,
-	inferenceProviderMapping?: InferenceProviderModelMapping,
+	inferenceProviderMapping?: InferenceProviderMappingEntry,
 	opts?: Record<string, unknown>
 ): InferenceSnippet[] {
 	return model.pipeline_tag && model.pipeline_tag in snippets
@@ -444,21 +455,24 @@ function replaceAccessTokenPlaceholder(
 	placeholder: string,
 	snippet: string,
 	language: InferenceSnippetLanguage,
-	provider: InferenceProviderOrPolicy
+	provider: InferenceProviderOrPolicy,
+	endpointUrl?: string
 ): string {
 	// If "opts.accessToken" is not set, the snippets are generated with a placeholder.
 	// Once snippets are rendered, we replace the placeholder with code to fetch the access token from an environment variable.
 
 	// Determine if HF_TOKEN or specific provider token should be used
 	const useHfToken =
-		provider == "hf-inference" || // hf-inference provider => use $HF_TOKEN
-		(!directRequest && // if explicit directRequest => use provider-specific token
-			(!snippet.includes("https://") || // no URL provided => using a client => use $HF_TOKEN
-				snippet.includes("https://router.huggingface.co"))); // explicit routed request => use $HF_TOKEN
-
+		!endpointUrl && // custom endpointUrl => use a generic API_TOKEN
+		(provider == "hf-inference" || // hf-inference provider => use $HF_TOKEN
+			(!directRequest && // if explicit directRequest => use provider-specific token
+				(snippet.includes("InferenceClient") || // using a client => use $HF_TOKEN
+					snippet.includes("https://router.huggingface.co")))); // explicit routed request => use $HF_TOKEN
 	const accessTokenEnvVar = useHfToken
 		? "HF_TOKEN" // e.g. routed request or hf-inference
-		: provider.toUpperCase().replace("-", "_") + "_API_KEY"; // e.g. "REPLICATE_API_KEY"
+		: endpointUrl
+		  ? "API_TOKEN"
+		  : provider.toUpperCase().replace("-", "_") + "_API_KEY"; // e.g. "REPLICATE_API_KEY"
 
 	// Replace the placeholder with the env variable
 	if (language === "sh") {

package/src/snippets/templates.exported.ts

@@ -11,14 +11,14 @@ export const templates: Record<string, Record<string, Record<string, string>>> =
       "zeroShotClassification": "async function query(data) {\n    const response = await fetch(\n\t\t\"{{ fullUrl }}\",\n        {\n            headers: {\n\t\t\t\tAuthorization: \"{{ authorizationHeader }}\",\n                \"Content-Type\": \"application/json\",\n{% if billTo %}\n                \"X-HF-Bill-To\": \"{{ billTo }}\",\n{% endif %}         },\n            method: \"POST\",\n            body: JSON.stringify(data),\n        }\n    );\n    const result = await response.json();\n    return result;\n}\n\nquery({\n    inputs: {{ providerInputs.asObj.inputs }},\n    parameters: { candidate_labels: [\"refund\", \"legal\", \"faq\"] }\n}).then((response) => {\n    console.log(JSON.stringify(response));\n});"
     },
     "huggingface.js": {
-      "basic": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst output = await client.{{ methodName }}({\n\tmodel: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n\tprovider: \"{{ provider }}\",\n}{% if billTo %}, {\n\tbillTo: \"{{ billTo }}\",\n}{% endif %});\n\nconsole.log(output);",
-      "basicAudio": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst data = fs.readFileSync({{inputs.asObj.inputs}});\n\nconst output = await client.{{ methodName }}({\n\tdata,\n\tmodel: \"{{ model.id }}\",\n\tprovider: \"{{ provider }}\",\n}{% if billTo %}, {\n\tbillTo: \"{{ billTo }}\",\n}{% endif %});\n\nconsole.log(output);",
-      "basicImage": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst data = fs.readFileSync({{inputs.asObj.inputs}});\n\nconst output = await client.{{ methodName }}({\n\tdata,\n\tmodel: \"{{ model.id }}\",\n\tprovider: \"{{ provider }}\",\n}{% if billTo %}, {\n\tbillTo: \"{{ billTo }}\",\n}{% endif %});\n\nconsole.log(output);",
-      "conversational": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst chatCompletion = await client.chatCompletion({\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n{{ inputs.asTsString }}\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n\nconsole.log(chatCompletion.choices[0].message);",
-      "conversationalStream": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nlet out = \"\";\n\nconst stream = client.chatCompletionStream({\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n{{ inputs.asTsString }}\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n\nfor await (const chunk of stream) {\n\tif (chunk.choices && chunk.choices.length > 0) {\n\t\tconst newContent = chunk.choices[0].delta.content;\n\t\tout += newContent;\n\t\tconsole.log(newContent);\n\t}\n}",
-      "textToImage": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst image = await client.textToImage({\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n\tparameters: { num_inference_steps: 5 },\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n/// Use the generated image (it's a Blob)",
-      "textToSpeech": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst audio = await client.textToSpeech({\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n// Use the generated audio (it's a Blob)",
-      "textToVideo": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst video = await client.textToVideo({\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n// Use the generated video (it's a Blob)"
+      "basic": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst output = await client.{{ methodName }}({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n\tmodel: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n\tprovider: \"{{ provider }}\",\n}{% if billTo %}, {\n\tbillTo: \"{{ billTo }}\",\n}{% endif %});\n\nconsole.log(output);",
+      "basicAudio": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst data = fs.readFileSync({{inputs.asObj.inputs}});\n\nconst output = await client.{{ methodName }}({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n\tdata,\n\tmodel: \"{{ model.id }}\",\n\tprovider: \"{{ provider }}\",\n}{% if billTo %}, {\n\tbillTo: \"{{ billTo }}\",\n}{% endif %});\n\nconsole.log(output);",
+      "basicImage": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst data = fs.readFileSync({{inputs.asObj.inputs}});\n\nconst output = await client.{{ methodName }}({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n\tdata,\n\tmodel: \"{{ model.id }}\",\n\tprovider: \"{{ provider }}\",\n}{% if billTo %}, {\n\tbillTo: \"{{ billTo }}\",\n}{% endif %});\n\nconsole.log(output);",
+      "conversational": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst chatCompletion = await client.chatCompletion({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n{{ inputs.asTsString }}\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n\nconsole.log(chatCompletion.choices[0].message);",
+      "conversationalStream": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nlet out = \"\";\n\nconst stream = client.chatCompletionStream({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n{{ inputs.asTsString }}\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n\nfor await (const chunk of stream) {\n\tif (chunk.choices && chunk.choices.length > 0) {\n\t\tconst newContent = chunk.choices[0].delta.content;\n\t\tout += newContent;\n\t\tconsole.log(newContent);\n\t}\n}",
+      "textToImage": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst image = await client.textToImage({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n\tparameters: { num_inference_steps: 5 },\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n/// Use the generated image (it's a Blob)",
+      "textToSpeech": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst audio = await client.textToSpeech({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n// Use the generated audio (it's a Blob)",
+      "textToVideo": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst video = await client.textToVideo({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n// Use the generated video (it's a Blob)"
     },
     "openai": {
       "conversational": "import { OpenAI } from \"openai\";\n\nconst client = new OpenAI({\n\tbaseURL: \"{{ baseUrl }}\",\n\tapiKey: \"{{ accessToken }}\",\n{% if billTo %}\n\tdefaultHeaders: {\n\t\t\"X-HF-Bill-To\": \"{{ billTo }}\" \n\t}\n{% endif %}\n});\n\nconst chatCompletion = await client.chat.completions.create({\n\tmodel: \"{{ providerModelId }}\",\n{{ inputs.asTsString }}\n});\n\nconsole.log(chatCompletion.choices[0].message);",
@@ -37,7 +37,7 @@ export const templates: Record<string, Record<string, Record<string, string>>> =
       "conversationalStream": "stream = client.chat.completions.create(\n    model=\"{{ model.id }}\",\n{{ inputs.asPythonString }}\n    stream=True,\n)\n\nfor chunk in stream:\n    print(chunk.choices[0].delta.content, end=\"\") ",
       "documentQuestionAnswering": "output = client.document_question_answering(\n    \"{{ inputs.asObj.image }}\",\n    question=\"{{ inputs.asObj.question }}\",\n    model=\"{{ model.id }}\",\n) ",
       "imageToImage": "# output is a PIL.Image object\nimage = client.image_to_image(\n    \"{{ inputs.asObj.inputs }}\",\n    prompt=\"{{ inputs.asObj.parameters.prompt }}\",\n    model=\"{{ model.id }}\",\n) ",
-      "importInferenceClient": "from huggingface_hub import InferenceClient\n\nclient = InferenceClient(\n    provider=\"{{ provider }}\",\n    api_key=\"{{ accessToken }}\",\n{% if billTo %}\n    bill_to=\"{{ billTo }}\",\n{% endif %}\n)",
+      "importInferenceClient": "from huggingface_hub import InferenceClient\n\nclient = InferenceClient(\n{% if endpointUrl %}\n    base_url=\"{{ baseUrl }}\",\n{% endif %}\n    provider=\"{{ provider }}\",\n    api_key=\"{{ accessToken }}\",\n{% if billTo %}\n    bill_to=\"{{ billTo }}\",\n{% endif %}\n)",
       "questionAnswering": "answer = client.question_answering(\n    question=\"{{ inputs.asObj.question }}\",\n    context=\"{{ inputs.asObj.context }}\",\n    model=\"{{ model.id }}\",\n) ",
       "tableQuestionAnswering": "answer = client.table_question_answering(\n    query=\"{{ inputs.asObj.query }}\",\n    table={{ inputs.asObj.table }},\n    model=\"{{ model.id }}\",\n) ",
       "textToImage": "# output is a PIL.Image object\nimage = client.text_to_image(\n    {{ inputs.asObj.inputs }},\n    model=\"{{ model.id }}\",\n) ",

package/src/types.ts

@@ -1,5 +1,5 @@
 import type { ChatCompletionInput, PipelineType } from "@huggingface/tasks";
-import type { InferenceProviderModelMapping } from "./lib/getInferenceProviderMapping.js";
+import type { InferenceProviderMappingEntry } from "./lib/getInferenceProviderMapping.js";
 
 /**
  * HF model id, like "meta-llama/Llama-3.3-70B-Instruct"
@@ -126,6 +126,6 @@ export interface UrlParams {
 export interface BodyParams<T extends Record<string, unknown> = Record<string, unknown>> {
 	args: T;
 	model: string;
-	mapping?: InferenceProviderModelMapping | undefined;
+	mapping?: InferenceProviderMappingEntry | undefined;
 	task?: InferenceTask;
 }