@huggingface/inference 4.0.0 → 4.0.1

package/dist/commonjs/package.d.ts

@@ -1,3 +1,3 @@
-export declare const PACKAGE_VERSION = "4.0.0";
+export declare const PACKAGE_VERSION = "4.0.1";
 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 = "4.0.0";
+exports.PACKAGE_VERSION = "4.0.1";
 exports.PACKAGE_NAME = "@huggingface/inference";

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

@@ -4,6 +4,7 @@ import type { InferenceProviderOrPolicy } from "../types.js";
 export type InferenceSnippetOptions = {
     streaming?: boolean;
     billTo?: string;
+    accessToken?: string;
 } & Record<string, unknown>;
-export declare function getInferenceSnippets(model: ModelDataMinimal, accessToken: string, provider: InferenceProviderOrPolicy, inferenceProviderMapping?: InferenceProviderModelMapping, opts?: Record<string, unknown>): InferenceSnippet[];
+export declare function getInferenceSnippets(model: ModelDataMinimal, provider: InferenceProviderOrPolicy, inferenceProviderMapping?: InferenceProviderModelMapping, 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;IAAE,SAAS,CAAC,EAAE,OAAO,CAAC;IAAC,MAAM,CAAC,EAAE,MAAM,CAAA;CAAE,GAAG,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;AAmUzG,wBAAgB,oBAAoB,CACnC,KAAK,EAAE,gBAAgB,EACvB,WAAW,EAAE,MAAM,EACnB,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;IAAE,SAAS,CAAC,EAAE,OAAO,CAAC;IAAC,MAAM,CAAC,EAAE,MAAM,CAAC;IAAC,WAAW,CAAC,EAAE,MAAM,CAAA;CAAE,GAAG,MAAM,CAC5G,MAAM,EACN,OAAO,CACP,CAAC;AA0UF,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

@@ -76,9 +76,10 @@ const HF_JS_METHODS = {
     "text-to-speech": "textToSpeech",
     translation: "translation",
 };
+const ACCESS_TOKEN_PLACEHOLDER = "<ACCESS_TOKEN>"; // Placeholder to replace with env variable in snippets
 // Snippet generators
 const snippetGenerator = (templateName, inputPreparationFn) => {
-    return (model, accessToken, provider, inferenceProviderMapping, opts) => {
+    return (model, provider, inferenceProviderMapping, opts) => {
         const providerModelId = inferenceProviderMapping?.providerId ?? model.id;
         /// Hacky: hard-code conversational templates here
         let task = model.pipeline_tag;
@@ -97,10 +98,11 @@ const snippetGenerator = (templateName, inputPreparationFn) => {
             console.error(`Failed to get provider helper for ${provider} (${task})`, e);
             return [];
         }
+        const accessTokenOrPlaceholder = opts?.accessToken ?? ACCESS_TOKEN_PLACEHOLDER;
         /// Prepare inputs + make request
         const inputs = inputPreparationFn ? inputPreparationFn(model, opts) : { inputs: (0, tasks_1.getModelInputSnippet)(model) };
         const request = (0, makeRequestOptions_js_1.makeRequestOptionsFromResolvedModel)(providerModelId, providerHelper, {
-            accessToken,
+            accessToken: accessTokenOrPlaceholder,
             provider,
             ...inputs,
         }, inferenceProviderMapping, {
@@ -121,7 +123,7 @@ const snippetGenerator = (templateName, inputPreparationFn) => {
         }
         /// Prepare template injection data
         const params = {
-            accessToken,
+            accessToken: accessTokenOrPlaceholder,
             authorizationHeader: request.info.headers?.Authorization,
             baseUrl: removeSuffix(request.url, "/chat/completions"),
             fullUrl: request.url,
@@ -185,6 +187,10 @@ const snippetGenerator = (templateName, inputPreparationFn) => {
                     });
                     snippet = `${importSection}\n\n${snippet}`;
                 }
+                /// Replace access token placeholder
+                if (snippet.includes(ACCESS_TOKEN_PLACEHOLDER)) {
+                    snippet = replaceAccessTokenPlaceholder(snippet, language, provider);
+                }
                 /// Snippet is ready!
                 return { language, client: client, content: snippet };
             })
@@ -247,9 +253,9 @@ const snippets = {
     "zero-shot-classification": snippetGenerator("zeroShotClassification"),
     "zero-shot-image-classification": snippetGenerator("zeroShotImageClassification"),
 };
-function getInferenceSnippets(model, accessToken, provider, inferenceProviderMapping, opts) {
+function getInferenceSnippets(model, provider, inferenceProviderMapping, opts) {
     return model.pipeline_tag && model.pipeline_tag in snippets
-        ? snippets[model.pipeline_tag]?.(model, accessToken, provider, inferenceProviderMapping, opts) ?? []
+        ? snippets[model.pipeline_tag]?.(model, provider, inferenceProviderMapping, opts) ?? []
         : [];
 }
 // String manipulation helpers
@@ -310,3 +316,36 @@ function indentString(str) {
 function removeSuffix(str, suffix) {
     return str.endsWith(suffix) ? str.slice(0, -suffix.length) : str;
 }
+function replaceAccessTokenPlaceholder(snippet, language, provider) {
+    // 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 accessTokenEnvVar = !snippet.includes("https://") || // no URL provided => using a client => use $HF_TOKEN
+        snippet.includes("https://router.huggingface.co") || // explicit routed request => use $HF_TOKEN
+        provider == "hf-inference" // hf-inference provider => use $HF_TOKEN
+        ? "HF_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 ${ACCESS_TOKEN_PLACEHOLDER}'`, `"Authorization: Bearer $${accessTokenEnvVar}"` // e.g. "Authorization: Bearer $HF_TOKEN"
+        );
+    }
+    else if (language === "python") {
+        snippet = "import os\n" + snippet;
+        snippet = snippet.replace(`"${ACCESS_TOKEN_PLACEHOLDER}"`, `os.environ["${accessTokenEnvVar}"]` // e.g. os.environ["HF_TOKEN")
+        );
+        snippet = snippet.replace(`"Bearer ${ACCESS_TOKEN_PLACEHOLDER}"`, `f"Bearer {os.environ['${accessTokenEnvVar}']}"` // e.g. f"Bearer {os.environ['HF_TOKEN']}"
+        );
+        snippet = snippet.replace(`"Key ${ACCESS_TOKEN_PLACEHOLDER}"`, `f"Key {os.environ['${accessTokenEnvVar}']}"` // e.g. f"Key {os.environ['FAL_AI_API_KEY']}"
+        );
+    }
+    else if (language === "js") {
+        snippet = snippet.replace(`"${ACCESS_TOKEN_PLACEHOLDER}"`, `process.env.${accessTokenEnvVar}` // e.g. process.env.HF_TOKEN
+        );
+        snippet = snippet.replace(`Authorization: "Bearer ${ACCESS_TOKEN_PLACEHOLDER}",`, `Authorization: \`Bearer $\{process.env.${accessTokenEnvVar}}\`,` // e.g. Authorization: `Bearer ${process.env.HF_TOKEN}`,
+        );
+        snippet = snippet.replace(`Authorization: "Key ${ACCESS_TOKEN_PLACEHOLDER}",`, `Authorization: \`Key $\{process.env.${accessTokenEnvVar}}\`,` // e.g. Authorization: `Key ${process.env.FAL_AI_API_KEY}`,
+        );
+    }
+    return snippet;
+}

package/dist/commonjs/utils/request.js

@@ -35,8 +35,8 @@ async function innerRequest(args, providerHelper, options) {
                     body: requestArgsToJson(args),
                 }, { requestId: response.headers.get("x-request-id") ?? "", status: response.status, body: output });
             }
-            if (typeof output.error === "string" || typeof output.detail === "string") {
-                throw new errors_js_1.InferenceClientProviderApiError(`Failed to perform inference: ${output.error ?? output.detail}`, {
+            if (typeof output.error === "string" || typeof output.detail === "string" || typeof output.message === "string") {
+                throw new errors_js_1.InferenceClientProviderApiError(`Failed to perform inference: ${output.error ?? output.detail ?? output.message}`, {
                     url,
                     method: info.method ?? "GET",
                     headers: info.headers,

package/dist/esm/package.d.ts

@@ -1,3 +1,3 @@
-export declare const PACKAGE_VERSION = "4.0.0";
+export declare const PACKAGE_VERSION = "4.0.1";
 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 = "4.0.0";
+export const PACKAGE_VERSION = "4.0.1";
 export const PACKAGE_NAME = "@huggingface/inference";

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

@@ -4,6 +4,7 @@ import type { InferenceProviderOrPolicy } from "../types.js";
 export type InferenceSnippetOptions = {
     streaming?: boolean;
     billTo?: string;
+    accessToken?: string;
 } & Record<string, unknown>;
-export declare function getInferenceSnippets(model: ModelDataMinimal, accessToken: string, provider: InferenceProviderOrPolicy, inferenceProviderMapping?: InferenceProviderModelMapping, opts?: Record<string, unknown>): InferenceSnippet[];
+export declare function getInferenceSnippets(model: ModelDataMinimal, provider: InferenceProviderOrPolicy, inferenceProviderMapping?: InferenceProviderModelMapping, 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;IAAE,SAAS,CAAC,EAAE,OAAO,CAAC;IAAC,MAAM,CAAC,EAAE,MAAM,CAAA;CAAE,GAAG,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;AAmUzG,wBAAgB,oBAAoB,CACnC,KAAK,EAAE,gBAAgB,EACvB,WAAW,EAAE,MAAM,EACnB,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;IAAE,SAAS,CAAC,EAAE,OAAO,CAAC;IAAC,MAAM,CAAC,EAAE,MAAM,CAAC;IAAC,WAAW,CAAC,EAAE,MAAM,CAAA;CAAE,GAAG,MAAM,CAC5G,MAAM,EACN,OAAO,CACP,CAAC;AA0UF,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

@@ -73,9 +73,10 @@ const HF_JS_METHODS = {
     "text-to-speech": "textToSpeech",
     translation: "translation",
 };
+const ACCESS_TOKEN_PLACEHOLDER = "<ACCESS_TOKEN>"; // Placeholder to replace with env variable in snippets
 // Snippet generators
 const snippetGenerator = (templateName, inputPreparationFn) => {
-    return (model, accessToken, provider, inferenceProviderMapping, opts) => {
+    return (model, provider, inferenceProviderMapping, opts) => {
         const providerModelId = inferenceProviderMapping?.providerId ?? model.id;
         /// Hacky: hard-code conversational templates here
         let task = model.pipeline_tag;
@@ -94,10 +95,11 @@ const snippetGenerator = (templateName, inputPreparationFn) => {
             console.error(`Failed to get provider helper for ${provider} (${task})`, e);
             return [];
         }
+        const accessTokenOrPlaceholder = opts?.accessToken ?? ACCESS_TOKEN_PLACEHOLDER;
         /// Prepare inputs + make request
         const inputs = inputPreparationFn ? inputPreparationFn(model, opts) : { inputs: getModelInputSnippet(model) };
         const request = makeRequestOptionsFromResolvedModel(providerModelId, providerHelper, {
-            accessToken,
+            accessToken: accessTokenOrPlaceholder,
             provider,
             ...inputs,
         }, inferenceProviderMapping, {
@@ -118,7 +120,7 @@ const snippetGenerator = (templateName, inputPreparationFn) => {
         }
         /// Prepare template injection data
         const params = {
-            accessToken,
+            accessToken: accessTokenOrPlaceholder,
             authorizationHeader: request.info.headers?.Authorization,
             baseUrl: removeSuffix(request.url, "/chat/completions"),
             fullUrl: request.url,
@@ -182,6 +184,10 @@ const snippetGenerator = (templateName, inputPreparationFn) => {
                     });
                     snippet = `${importSection}\n\n${snippet}`;
                 }
+                /// Replace access token placeholder
+                if (snippet.includes(ACCESS_TOKEN_PLACEHOLDER)) {
+                    snippet = replaceAccessTokenPlaceholder(snippet, language, provider);
+                }
                 /// Snippet is ready!
                 return { language, client: client, content: snippet };
             })
@@ -244,9 +250,9 @@ const snippets = {
     "zero-shot-classification": snippetGenerator("zeroShotClassification"),
     "zero-shot-image-classification": snippetGenerator("zeroShotImageClassification"),
 };
-export function getInferenceSnippets(model, accessToken, provider, inferenceProviderMapping, opts) {
+export function getInferenceSnippets(model, provider, inferenceProviderMapping, opts) {
     return model.pipeline_tag && model.pipeline_tag in snippets
-        ? snippets[model.pipeline_tag]?.(model, accessToken, provider, inferenceProviderMapping, opts) ?? []
+        ? snippets[model.pipeline_tag]?.(model, provider, inferenceProviderMapping, opts) ?? []
         : [];
 }
 // String manipulation helpers
@@ -307,3 +313,36 @@ function indentString(str) {
 function removeSuffix(str, suffix) {
     return str.endsWith(suffix) ? str.slice(0, -suffix.length) : str;
 }
+function replaceAccessTokenPlaceholder(snippet, language, provider) {
+    // 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 accessTokenEnvVar = !snippet.includes("https://") || // no URL provided => using a client => use $HF_TOKEN
+        snippet.includes("https://router.huggingface.co") || // explicit routed request => use $HF_TOKEN
+        provider == "hf-inference" // hf-inference provider => use $HF_TOKEN
+        ? "HF_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 ${ACCESS_TOKEN_PLACEHOLDER}'`, `"Authorization: Bearer $${accessTokenEnvVar}"` // e.g. "Authorization: Bearer $HF_TOKEN"
+        );
+    }
+    else if (language === "python") {
+        snippet = "import os\n" + snippet;
+        snippet = snippet.replace(`"${ACCESS_TOKEN_PLACEHOLDER}"`, `os.environ["${accessTokenEnvVar}"]` // e.g. os.environ["HF_TOKEN")
+        );
+        snippet = snippet.replace(`"Bearer ${ACCESS_TOKEN_PLACEHOLDER}"`, `f"Bearer {os.environ['${accessTokenEnvVar}']}"` // e.g. f"Bearer {os.environ['HF_TOKEN']}"
+        );
+        snippet = snippet.replace(`"Key ${ACCESS_TOKEN_PLACEHOLDER}"`, `f"Key {os.environ['${accessTokenEnvVar}']}"` // e.g. f"Key {os.environ['FAL_AI_API_KEY']}"
+        );
+    }
+    else if (language === "js") {
+        snippet = snippet.replace(`"${ACCESS_TOKEN_PLACEHOLDER}"`, `process.env.${accessTokenEnvVar}` // e.g. process.env.HF_TOKEN
+        );
+        snippet = snippet.replace(`Authorization: "Bearer ${ACCESS_TOKEN_PLACEHOLDER}",`, `Authorization: \`Bearer $\{process.env.${accessTokenEnvVar}}\`,` // e.g. Authorization: `Bearer ${process.env.HF_TOKEN}`,
+        );
+        snippet = snippet.replace(`Authorization: "Key ${ACCESS_TOKEN_PLACEHOLDER}",`, `Authorization: \`Key $\{process.env.${accessTokenEnvVar}}\`,` // e.g. Authorization: `Key ${process.env.FAL_AI_API_KEY}`,
+        );
+    }
+    return snippet;
+}

package/dist/esm/utils/request.js

@@ -31,8 +31,8 @@ export async function innerRequest(args, providerHelper, options) {
                     body: requestArgsToJson(args),
                 }, { requestId: response.headers.get("x-request-id") ?? "", status: response.status, body: output });
             }
-            if (typeof output.error === "string" || typeof output.detail === "string") {
-                throw new InferenceClientProviderApiError(`Failed to perform inference: ${output.error ?? output.detail}`, {
+            if (typeof output.error === "string" || typeof output.detail === "string" || typeof output.message === "string") {
+                throw new InferenceClientProviderApiError(`Failed to perform inference: ${output.error ?? output.detail ?? output.message}`, {
                     url,
                     method: info.method ?? "GET",
                     headers: info.headers,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@huggingface/inference",
-  "version": "4.0.0",
+  "version": "4.0.1",
   "license": "MIT",
   "author": "Hugging Face and Tim Mikeladze <tim.mikeladze@gmail.com>",
   "description": "Typescript client for the Hugging Face Inference Providers and Inference Endpoints",

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 = "4.0.0";
+export const PACKAGE_VERSION = "4.0.1";
 export const PACKAGE_NAME = "@huggingface/inference";

package/src/snippets/getInferenceSnippets.ts

@@ -14,7 +14,10 @@ import { makeRequestOptionsFromResolvedModel } from "../lib/makeRequestOptions.j
 import type { InferenceProviderOrPolicy, InferenceTask, RequestArgs } from "../types.js";
 import { templates } from "./templates.exported.js";
 
-export type InferenceSnippetOptions = { streaming?: boolean; billTo?: string } & Record<string, unknown>;
+export type InferenceSnippetOptions = { streaming?: boolean; billTo?: string; accessToken?: string } & Record<
+	string,
+	unknown
+>;
 
 const PYTHON_CLIENTS = ["huggingface_hub", "fal_client", "requests", "openai"] as const;
 const JS_CLIENTS = ["fetch", "huggingface.js", "openai"] as const;
@@ -121,11 +124,12 @@ const HF_JS_METHODS: Partial<Record<WidgetType, string>> = {
 	translation: "translation",
 };
 
+const ACCESS_TOKEN_PLACEHOLDER = "<ACCESS_TOKEN>"; // Placeholder to replace with env variable in snippets
+
 // Snippet generators
 const snippetGenerator = (templateName: string, inputPreparationFn?: InputPreparationFn) => {
 	return (
 		model: ModelDataMinimal,
-		accessToken: string,
 		provider: InferenceProviderOrPolicy,
 		inferenceProviderMapping?: InferenceProviderModelMapping,
 		opts?: InferenceSnippetOptions
@@ -149,13 +153,15 @@ const snippetGenerator = (templateName: string, inputPreparationFn?: InputPrepar
 			console.error(`Failed to get provider helper for ${provider} (${task})`, e);
 			return [];
 		}
+		const accessTokenOrPlaceholder = opts?.accessToken ?? ACCESS_TOKEN_PLACEHOLDER;
+
 		/// Prepare inputs + make request
 		const inputs = inputPreparationFn ? inputPreparationFn(model, opts) : { inputs: getModelInputSnippet(model) };
 		const request = makeRequestOptionsFromResolvedModel(
 			providerModelId,
 			providerHelper,
 			{
-				accessToken,
+				accessToken: accessTokenOrPlaceholder,
 				provider,
 				...inputs,
 			} as RequestArgs,
@@ -180,7 +186,7 @@ const snippetGenerator = (templateName: string, inputPreparationFn?: InputPrepar
 
 		/// Prepare template injection data
 		const params: TemplateParams = {
-			accessToken,
+			accessToken: accessTokenOrPlaceholder,
 			authorizationHeader: (request.info.headers as Record<string, string>)?.Authorization,
 			baseUrl: removeSuffix(request.url, "/chat/completions"),
 			fullUrl: request.url,
@@ -248,6 +254,11 @@ const snippetGenerator = (templateName: string, inputPreparationFn?: InputPrepar
 							snippet = `${importSection}\n\n${snippet}`;
 						}
 
+						/// Replace access token placeholder
+						if (snippet.includes(ACCESS_TOKEN_PLACEHOLDER)) {
+							snippet = replaceAccessTokenPlaceholder(snippet, language, provider);
+						}
+
 						/// Snippet is ready!
 						return { language, client: client as string, content: snippet };
 					})
@@ -299,7 +310,6 @@ const snippets: Partial<
 		PipelineType,
 		(
 			model: ModelDataMinimal,
-			accessToken: string,
 			provider: InferenceProviderOrPolicy,
 			inferenceProviderMapping?: InferenceProviderModelMapping,
 			opts?: InferenceSnippetOptions
@@ -339,13 +349,12 @@ const snippets: Partial<
 
 export function getInferenceSnippets(
 	model: ModelDataMinimal,
-	accessToken: string,
 	provider: InferenceProviderOrPolicy,
 	inferenceProviderMapping?: InferenceProviderModelMapping,
 	opts?: Record<string, unknown>
 ): InferenceSnippet[] {
 	return model.pipeline_tag && model.pipeline_tag in snippets
-		? snippets[model.pipeline_tag]?.(model, accessToken, provider, inferenceProviderMapping, opts) ?? []
+		? snippets[model.pipeline_tag]?.(model, provider, inferenceProviderMapping, opts) ?? []
 		: [];
 }
 
@@ -420,3 +429,56 @@ function indentString(str: string): string {
 function removeSuffix(str: string, suffix: string) {
 	return str.endsWith(suffix) ? str.slice(0, -suffix.length) : str;
 }
+
+function replaceAccessTokenPlaceholder(
+	snippet: string,
+	language: InferenceSnippetLanguage,
+	provider: InferenceProviderOrPolicy
+): 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 accessTokenEnvVar =
+		!snippet.includes("https://") || // no URL provided => using a client => use $HF_TOKEN
+		snippet.includes("https://router.huggingface.co") || // explicit routed request => use $HF_TOKEN
+		provider == "hf-inference" // hf-inference provider => use $HF_TOKEN
+			? "HF_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 ${ACCESS_TOKEN_PLACEHOLDER}'`,
+			`"Authorization: Bearer $${accessTokenEnvVar}"` // e.g. "Authorization: Bearer $HF_TOKEN"
+		);
+	} else if (language === "python") {
+		snippet = "import os\n" + snippet;
+		snippet = snippet.replace(
+			`"${ACCESS_TOKEN_PLACEHOLDER}"`,
+			`os.environ["${accessTokenEnvVar}"]` // e.g. os.environ["HF_TOKEN")
+		);
+		snippet = snippet.replace(
+			`"Bearer ${ACCESS_TOKEN_PLACEHOLDER}"`,
+			`f"Bearer {os.environ['${accessTokenEnvVar}']}"` // e.g. f"Bearer {os.environ['HF_TOKEN']}"
+		);
+		snippet = snippet.replace(
+			`"Key ${ACCESS_TOKEN_PLACEHOLDER}"`,
+			`f"Key {os.environ['${accessTokenEnvVar}']}"` // e.g. f"Key {os.environ['FAL_AI_API_KEY']}"
+		);
+	} else if (language === "js") {
+		snippet = snippet.replace(
+			`"${ACCESS_TOKEN_PLACEHOLDER}"`,
+			`process.env.${accessTokenEnvVar}` // e.g. process.env.HF_TOKEN
+		);
+		snippet = snippet.replace(
+			`Authorization: "Bearer ${ACCESS_TOKEN_PLACEHOLDER}",`,
+			`Authorization: \`Bearer $\{process.env.${accessTokenEnvVar}}\`,` // e.g. Authorization: `Bearer ${process.env.HF_TOKEN}`,
+		);
+		snippet = snippet.replace(
+			`Authorization: "Key ${ACCESS_TOKEN_PLACEHOLDER}",`,
+			`Authorization: \`Key $\{process.env.${accessTokenEnvVar}}\`,` // e.g. Authorization: `Key ${process.env.FAL_AI_API_KEY}`,
+		);
+	}
+	return snippet;
+}

package/src/utils/request.ts

@@ -65,9 +65,9 @@ export async function innerRequest<T>(
 					{ requestId: response.headers.get("x-request-id") ?? "", status: response.status, body: output }
 				);
 			}
-			if (typeof output.error === "string" || typeof output.detail === "string") {
+			if (typeof output.error === "string" || typeof output.detail === "string" || typeof output.message === "string") {
 				throw new InferenceClientProviderApiError(
-					`Failed to perform inference: ${output.error ?? output.detail}`,
+					`Failed to perform inference: ${output.error ?? output.detail ?? output.message}`,
 					{
 						url,
 						method: info.method ?? "GET",