@huggingface/inference 3.5.1 → 3.6.0

package/dist/src/{HfInference.d.ts → InferenceClient.d.ts}

@@ -8,21 +8,26 @@ type TaskWithNoAccessToken = {
 type TaskWithNoAccessTokenNoEndpointUrl = {
     [key in keyof Task]: (args: DistributiveOmit<Parameters<Task[key]>[0], "accessToken" | "endpointUrl">, options?: Parameters<Task[key]>[1]) => ReturnType<Task[key]>;
 };
-export declare class HfInference {
+export declare class InferenceClient {
     private readonly accessToken;
     private readonly defaultOptions;
     constructor(accessToken?: string, defaultOptions?: Options);
     /**
-     * Returns copy of HfInference tied to a specified endpoint.
+     * Returns copy of InferenceClient tied to a specified endpoint.
      */
-    endpoint(endpointUrl: string): HfInferenceEndpoint;
+    endpoint(endpointUrl: string): InferenceClientEndpoint;
 }
-export declare class HfInferenceEndpoint {
+export declare class InferenceClientEndpoint {
     constructor(endpointUrl: string, accessToken?: string, defaultOptions?: Options);
 }
-export interface HfInference extends TaskWithNoAccessToken {
+export interface InferenceClient extends TaskWithNoAccessToken {
 }
-export interface HfInferenceEndpoint extends TaskWithNoAccessTokenNoEndpointUrl {
+export interface InferenceClientEndpoint extends TaskWithNoAccessTokenNoEndpointUrl {
+}
+/**
+ * For backward compatibility only.
+ */
+export declare class HfInference extends InferenceClient {
 }
 export {};
-//# sourceMappingURL=HfInference.d.ts.map
+//# sourceMappingURL=InferenceClient.d.ts.map

package/dist/src/InferenceClient.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"InferenceClient.d.ts","sourceRoot":"","sources":["../../src/InferenceClient.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,KAAK,MAAM,SAAS,CAAC;AACjC,OAAO,KAAK,EAAE,OAAO,EAAe,MAAM,SAAS,CAAC;AACpD,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,2BAA2B,CAAC;AAKlE,KAAK,IAAI,GAAG,OAAO,KAAK,CAAC;AAEzB,KAAK,qBAAqB,GAAG;KAC3B,GAAG,IAAI,MAAM,IAAI,GAAG,CACpB,IAAI,EAAE,gBAAgB,CAAC,UAAU,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,aAAa,CAAC,EAC/D,OAAO,CAAC,EAAE,UAAU,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,KAC9B,UAAU,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;CAC1B,CAAC;AAEF,KAAK,kCAAkC,GAAG;KACxC,GAAG,IAAI,MAAM,IAAI,GAAG,CACpB,IAAI,EAAE,gBAAgB,CAAC,UAAU,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,aAAa,GAAG,aAAa,CAAC,EAC/E,OAAO,CAAC,EAAE,UAAU,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,KAC9B,UAAU,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;CAC1B,CAAC;AAEF,qBAAa,eAAe;IAC3B,OAAO,CAAC,QAAQ,CAAC,WAAW,CAAS;IACrC,OAAO,CAAC,QAAQ,CAAC,cAAc,CAAU;gBAE7B,WAAW,SAAK,EAAE,cAAc,GAAE,OAAY;IAc1D;;OAEG;IACI,QAAQ,CAAC,WAAW,EAAE,MAAM,GAAG,uBAAuB;CAG7D;AAED,qBAAa,uBAAuB;gBACvB,WAAW,EAAE,MAAM,EAAE,WAAW,SAAK,EAAE,cAAc,GAAE,OAAY;CAa/E;AAED,MAAM,WAAW,eAAgB,SAAQ,qBAAqB;CAAG;AAEjE,MAAM,WAAW,uBAAwB,SAAQ,kCAAkC;CAAG;AAEtF;;GAEG;AACH,qBAAa,WAAY,SAAQ,eAAe;CAAG"}

package/dist/src/index.d.ts

@@ -1,4 +1,4 @@
-export { HfInference, HfInferenceEndpoint } from "./HfInference";
+export { InferenceClient, InferenceClientEndpoint, HfInference } from "./InferenceClient";
 export { InferenceOutputError } from "./lib/InferenceOutputError";
 export * from "./types";
 export * from "./tasks";

package/dist/src/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,mBAAmB,EAAE,MAAM,eAAe,CAAC;AACjE,OAAO,EAAE,oBAAoB,EAAE,MAAM,4BAA4B,CAAC;AAClE,cAAc,SAAS,CAAC;AACxB,cAAc,SAAS,CAAC;AAExB,OAAO,KAAK,QAAQ,MAAM,qBAAqB,CAAC;AAChD,OAAO,EAAE,QAAQ,EAAE,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,eAAe,EAAE,uBAAuB,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAC1F,OAAO,EAAE,oBAAoB,EAAE,MAAM,4BAA4B,CAAC;AAClE,cAAc,SAAS,CAAC;AACxB,cAAc,SAAS,CAAC;AACxB,OAAO,KAAK,QAAQ,MAAM,qBAAqB,CAAC;AAEhD,OAAO,EAAE,QAAQ,EAAE,CAAC"}

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

@@ -1,6 +1,7 @@
 import type { InferenceTask, Options, RequestArgs } from "../types";
 /**
- * Helper that prepares request arguments
+ * Helper that prepares request arguments.
+ * This async version handle the model ID resolution step.
  */
 export declare function makeRequestOptions(args: RequestArgs & {
     data?: Blob | ArrayBuffer;
@@ -13,4 +14,18 @@ export declare function makeRequestOptions(args: RequestArgs & {
     url: string;
     info: RequestInit;
 }>;
+/**
+ * Helper that prepares request arguments. - for internal use only
+ * This sync version skips the model ID resolution step
+ */
+export declare function makeRequestOptionsFromResolvedModel(resolvedModel: string, args: RequestArgs & {
+    data?: Blob | ArrayBuffer;
+    stream?: boolean;
+}, options?: Options & {
+    task?: InferenceTask;
+    chatCompletion?: boolean;
+}): {
+    url: string;
+    info: RequestInit;
+};
 //# sourceMappingURL=makeRequestOptions.d.ts.map

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

@@ -1 +1 @@
-{"version":3,"file":"makeRequestOptions.d.ts","sourceRoot":"","sources":["../../../src/lib/makeRequestOptions.ts"],"names":[],"mappings":"AAcA,OAAO,KAAK,EAAqB,aAAa,EAAE,OAAO,EAAkB,WAAW,EAAE,MAAM,UAAU,CAAC;AAgCvG;;GAEG;AACH,wBAAsB,kBAAkB,CACvC,IAAI,EAAE,WAAW,GAAG;IACnB,IAAI,CAAC,EAAE,IAAI,GAAG,WAAW,CAAC;IAC1B,MAAM,CAAC,EAAE,OAAO,CAAC;CACjB,EACD,OAAO,CAAC,EAAE,OAAO,GAAG;IACnB,oEAAoE;IACpE,IAAI,CAAC,EAAE,aAAa,CAAC;IACrB,cAAc,CAAC,EAAE,OAAO,CAAC;CACzB,GACC,OAAO,CAAC;IAAE,GAAG,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,WAAW,CAAA;CAAE,CAAC,CAsH7C"}
+{"version":3,"file":"makeRequestOptions.d.ts","sourceRoot":"","sources":["../../../src/lib/makeRequestOptions.ts"],"names":[],"mappings":"AAcA,OAAO,KAAK,EAAqB,aAAa,EAAE,OAAO,EAAkB,WAAW,EAAE,MAAM,UAAU,CAAC;AAgCvG;;;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,OAAO,CAAC,EAAE,OAAO,GAAG;IACnB,oEAAoE;IACpE,IAAI,CAAC,EAAE,aAAa,CAAC;IACrB,cAAc,CAAC,EAAE,OAAO,CAAC;CACzB,GACC,OAAO,CAAC;IAAE,GAAG,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,WAAW,CAAA;CAAE,CAAC,CAoC7C;AAED;;;GAGG;AACH,wBAAgB,mCAAmC,CAClD,aAAa,EAAE,MAAM,EACrB,IAAI,EAAE,WAAW,GAAG;IACnB,IAAI,CAAC,EAAE,IAAI,GAAG,WAAW,CAAC;IAC1B,MAAM,CAAC,EAAE,OAAO,CAAC;CACjB,EACD,OAAO,CAAC,EAAE,OAAO,GAAG;IACnB,IAAI,CAAC,EAAE,aAAa,CAAC;IACrB,cAAc,CAAC,EAAE,OAAO,CAAC;CACzB,GACC;IAAE,GAAG,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,WAAW,CAAA;CAAE,CA4FpC"}

package/dist/src/providers/cohere.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"cohere.d.ts","sourceRoot":"","sources":["../../../src/providers/cohere.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AACH,OAAO,KAAK,EAAE,cAAc,EAAuC,MAAM,UAAU,CAAC;AAoBpF,eAAO,MAAM,aAAa,EAAE,cAK3B,CAAC"}
+{"version":3,"file":"cohere.d.ts","sourceRoot":"","sources":["../../../src/providers/cohere.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AACH,OAAO,KAAK,EAAE,cAAc,EAAuC,MAAM,UAAU,CAAC;AAmBpF,eAAO,MAAM,aAAa,EAAE,cAK3B,CAAC"}

package/dist/src/providers/novita.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"novita.d.ts","sourceRoot":"","sources":["../../../src/providers/novita.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AACH,OAAO,KAAK,EAAE,cAAc,EAAuC,MAAM,UAAU,CAAC;AAyBpF,eAAO,MAAM,aAAa,EAAE,cAK3B,CAAC"}
+{"version":3,"file":"novita.d.ts","sourceRoot":"","sources":["../../../src/providers/novita.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AACH,OAAO,KAAK,EAAE,cAAc,EAAuC,MAAM,UAAU,CAAC;AA0BpF,eAAO,MAAM,aAAa,EAAE,cAK3B,CAAC"}

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

@@ -0,0 +1,4 @@
+import { type InferenceSnippet, type ModelDataMinimal } from "@huggingface/tasks";
+import type { InferenceProvider } from "../types";
+export declare function getInferenceSnippets(model: ModelDataMinimal, accessToken: string, provider: InferenceProvider, providerModelId?: string, opts?: Record<string, unknown>): InferenceSnippet[];
+//# sourceMappingURL=getInferenceSnippets.d.ts.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"getInferenceSnippets.d.ts","sourceRoot":"","sources":["../../../src/snippets/getInferenceSnippets.ts"],"names":[],"mappings":"AAEA,OAAO,EACN,KAAK,gBAAgB,EAErB,KAAK,gBAAgB,EAGrB,MAAM,oBAAoB,CAAC;AAC5B,OAAO,KAAK,EAAE,iBAAiB,EAA8B,MAAM,UAAU,CAAC;AAkT9E,wBAAgB,oBAAoB,CACnC,KAAK,EAAE,gBAAgB,EACvB,WAAW,EAAE,MAAM,EACnB,QAAQ,EAAE,iBAAiB,EAC3B,eAAe,CAAC,EAAE,MAAM,EACxB,IAAI,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAC5B,gBAAgB,EAAE,CAIpB"}

package/dist/src/snippets/index.d.ts

@@ -1,5 +1,2 @@
-import * as curl from "./curl.js";
-import * as python from "./python.js";
-import * as js from "./js.js";
-export { curl, python, js };
+export { getInferenceSnippets } from "./getInferenceSnippets.js";
 //# sourceMappingURL=index.d.ts.map

package/dist/src/snippets/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/snippets/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,IAAI,MAAM,WAAW,CAAC;AAClC,OAAO,KAAK,MAAM,MAAM,aAAa,CAAC;AACtC,OAAO,KAAK,EAAE,MAAM,SAAS,CAAC;AAE9B,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,EAAE,EAAE,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/snippets/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,oBAAoB,EAAE,MAAM,2BAA2B,CAAC"}

package/dist/src/tasks/cv/textToVideo.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"textToVideo.d.ts","sourceRoot":"","sources":["../../../../src/tasks/cv/textToVideo.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,QAAQ,EAAqB,OAAO,EAAE,MAAM,aAAa,CAAC;AACxE,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,oBAAoB,CAAC;AAO3D,MAAM,MAAM,eAAe,GAAG,QAAQ,GAAG,gBAAgB,CAAC;AAE1D,MAAM,MAAM,iBAAiB,GAAG,IAAI,CAAC;AAcrC,wBAAsB,WAAW,CAAC,IAAI,EAAE,eAAe,EAAE,OAAO,CAAC,EAAE,OAAO,GAAG,OAAO,CAAC,iBAAiB,CAAC,CA0CtG"}
+{"version":3,"file":"textToVideo.d.ts","sourceRoot":"","sources":["../../../../src/tasks/cv/textToVideo.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,QAAQ,EAAqB,OAAO,EAAE,MAAM,aAAa,CAAC;AACxE,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,oBAAoB,CAAC;AAO3D,MAAM,MAAM,eAAe,GAAG,QAAQ,GAAG,gBAAgB,CAAC;AAE1D,MAAM,MAAM,iBAAiB,GAAG,IAAI,CAAC;AAoBrC,wBAAsB,WAAW,CAAC,IAAI,EAAE,eAAe,EAAE,OAAO,CAAC,EAAE,OAAO,GAAG,OAAO,CAAC,iBAAiB,CAAC,CAwDtG"}

package/dist/test/InferenceClient.spec.d.ts

@@ -0,0 +1,2 @@
+import "./vcr";
+//# sourceMappingURL=InferenceClient.spec.d.ts.map

package/dist/test/InferenceClient.spec.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"InferenceClient.spec.d.ts","sourceRoot":"","sources":["../../test/InferenceClient.spec.ts"],"names":[],"mappings":"AAeA,OAAO,OAAO,CAAC"}

package/package.json

@@ -1,10 +1,10 @@
 {
   "name": "@huggingface/inference",
-  "version": "3.5.1",
+  "version": "3.6.0",
   "packageManager": "pnpm@8.10.5",
   "license": "MIT",
-  "author": "Tim Mikeladze <tim.mikeladze@gmail.com>",
-  "description": "Typescript wrapper for the Hugging Face Inference Endpoints & Inference API",
+  "author": "Hugging Face and Tim Mikeladze <tim.mikeladze@gmail.com>",
+  "description": "Typescript client for the Hugging Face Inference Providers and Inference Endpoints",
   "repository": {
     "type": "git",
     "url": "https://github.com/huggingface/huggingface.js.git"
@@ -33,13 +33,21 @@
   "main": "./dist/index.cjs",
   "module": "./dist/index.js",
   "exports": {
-    "types": "./dist/src/index.d.ts",
-    "require": "./dist/index.cjs",
-    "import": "./dist/index.js"
+    ".": {
+      "types": "./dist/src/index.d.ts",
+      "require": "./dist/index.cjs",
+      "import": "./dist/index.js"
+    }
+  },
+  "browser": {
+    "./src/snippets/index.js": false,
+    "./dist/index.js": "./dist/browser/index.js",
+    "./dist/index.mjs": "./dist/browser/index.mjs"
   },
   "type": "module",
   "dependencies": {
-    "@huggingface/tasks": "^0.17.0"
+    "@huggingface/tasks": "^0.17.8",
+    "@huggingface/jinja": "^0.3.3"
   },
   "devDependencies": {
     "@types/node": "18.13.0"
@@ -54,6 +62,7 @@
     "format:check": "prettier --check .",
     "test": "vitest run --config vitest.config.mts",
     "test:browser": "vitest run --browser.name=chrome --browser.headless --config vitest.config.mts",
-    "check": "tsc"
+    "check": "tsc",
+    "dev": "tsup src/index.ts --format cjs,esm --watch"
   }
 }

package/src/{HfInference.ts → InferenceClient.ts}

@@ -21,7 +21,7 @@ type TaskWithNoAccessTokenNoEndpointUrl = {
 	) => ReturnType<Task[key]>;
 };
 
-export class HfInference {
+export class InferenceClient {
 	private readonly accessToken: string;
 	private readonly defaultOptions: Options;
 
@@ -40,14 +40,14 @@ export class HfInference {
 	}
 
 	/**
-	 * Returns copy of HfInference tied to a specified endpoint.
+	 * Returns copy of InferenceClient tied to a specified endpoint.
 	 */
-	public endpoint(endpointUrl: string): HfInferenceEndpoint {
-		return new HfInferenceEndpoint(endpointUrl, this.accessToken, this.defaultOptions);
+	public endpoint(endpointUrl: string): InferenceClientEndpoint {
+		return new InferenceClientEndpoint(endpointUrl, this.accessToken, this.defaultOptions);
 	}
 }
 
-export class HfInferenceEndpoint {
+export class InferenceClientEndpoint {
 	constructor(endpointUrl: string, accessToken = "", defaultOptions: Options = {}) {
 		accessToken;
 		defaultOptions;
@@ -63,6 +63,11 @@ export class HfInferenceEndpoint {
 	}
 }
 
-export interface HfInference extends TaskWithNoAccessToken {}
+export interface InferenceClient extends TaskWithNoAccessToken {}
 
-export interface HfInferenceEndpoint extends TaskWithNoAccessTokenNoEndpointUrl {}
+export interface InferenceClientEndpoint extends TaskWithNoAccessTokenNoEndpointUrl {}
+
+/**
+ * For backward compatibility only.
+ */
+export class HfInference extends InferenceClient {}

package/src/index.ts

@@ -1,7 +1,7 @@
-export { HfInference, HfInferenceEndpoint } from "./HfInference";
+export { InferenceClient, InferenceClientEndpoint, HfInference } from "./InferenceClient";
 export { InferenceOutputError } from "./lib/InferenceOutputError";
 export * from "./types";
 export * from "./tasks";
-
 import * as snippets from "./snippets/index.js";
+
 export { snippets };

package/src/lib/makeRequestOptions.ts

@@ -45,7 +45,8 @@ const providerConfigs: Record<InferenceProvider, ProviderConfig> = {
 };
 
 /**
- * Helper that prepares request arguments
+ * Helper that prepares request arguments.
+ * This async version handle the model ID resolution step.
  */
 export async function makeRequestOptions(
 	args: RequestArgs & {
@@ -58,13 +59,13 @@ export async function makeRequestOptions(
 		chatCompletion?: boolean;
 	}
 ): Promise<{ url: string; info: RequestInit }> {
-	const { accessToken, endpointUrl, provider: maybeProvider, model: maybeModel, ...remainingArgs } = args;
+	const { provider: maybeProvider, model: maybeModel } = args;
 	const provider = maybeProvider ?? "hf-inference";
 	const providerConfig = providerConfigs[provider];
+	const { task, chatCompletion } = options ?? {};
 
-	const { includeCredentials, task, chatCompletion, signal } = options ?? {};
-
-	if (endpointUrl && provider !== "hf-inference") {
+	// Validate inputs
+	if (args.endpointUrl && provider !== "hf-inference") {
 		throw new Error(`Cannot use endpointUrl with a third-party provider.`);
 	}
 	if (maybeModel && isUrl(maybeModel)) {
@@ -79,18 +80,44 @@ export async function makeRequestOptions(
 	if (providerConfig.clientSideRoutingOnly && !maybeModel) {
 		throw new Error(`Provider ${provider} requires a model ID to be passed directly.`);
 	}
+
 	// eslint-disable-next-line @typescript-eslint/no-non-null-assertion
 	const hfModel = maybeModel ?? (await loadDefaultModel(task!));
-	const model = providerConfig.clientSideRoutingOnly
+	const resolvedModel = providerConfig.clientSideRoutingOnly
 		? // eslint-disable-next-line @typescript-eslint/no-non-null-assertion
 		  removeProviderPrefix(maybeModel!, provider)
-		: // For closed-models API providers, one needs to pass the model ID directly (e.g. "gpt-3.5-turbo")
-		  await getProviderModelId({ model: hfModel, provider }, args, {
+		: await getProviderModelId({ model: hfModel, provider }, args, {
 				task,
 				chatCompletion,
 				fetch: options?.fetch,
 		  });
 
+	// Use the sync version with the resolved model
+	return makeRequestOptionsFromResolvedModel(resolvedModel, args, options);
+}
+
+/**
+ * Helper that prepares request arguments. - for internal use only
+ * This sync version skips the model ID resolution step
+ */
+export function makeRequestOptionsFromResolvedModel(
+	resolvedModel: string,
+	args: RequestArgs & {
+		data?: Blob | ArrayBuffer;
+		stream?: boolean;
+	},
+	options?: Options & {
+		task?: InferenceTask;
+		chatCompletion?: boolean;
+	}
+): { url: string; info: RequestInit } {
+	const { accessToken, endpointUrl, provider: maybeProvider, model, ...remainingArgs } = args;
+
+	const provider = maybeProvider ?? "hf-inference";
+	const providerConfig = providerConfigs[provider];
+
+	const { includeCredentials, task, chatCompletion, signal } = options ?? {};
+
 	const authMethod = (() => {
 		if (providerConfig.clientSideRoutingOnly) {
 			// Closed-source providers require an accessToken (cannot be routed).
@@ -119,7 +146,7 @@ export async function makeRequestOptions(
 					authMethod !== "provider-key"
 						? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", provider)
 						: providerConfig.baseUrl,
-				model,
+				model: resolvedModel,
 				chatCompletion,
 				task,
 		  });
@@ -150,7 +177,7 @@ export async function makeRequestOptions(
 		: JSON.stringify(
 				providerConfig.makeBody({
 					args: remainingArgs as Record<string, unknown>,
-					model,
+					model: resolvedModel,
 					task,
 					chatCompletion,
 				})

package/src/providers/black-forest-labs.ts

@@ -16,7 +16,7 @@
  */
 import type { ProviderConfig, UrlParams, HeaderParams, BodyParams } from "../types";
 
-const BLACK_FOREST_LABS_AI_API_BASE_URL = "https://api.us1.bfl.ai/v1";
+const BLACK_FOREST_LABS_AI_API_BASE_URL = "https://api.us1.bfl.ai";
 
 const makeBody = (params: BodyParams): Record<string, unknown> => {
 	return params.args;
@@ -31,7 +31,7 @@ const makeHeaders = (params: HeaderParams): Record<string, string> => {
 };
 
 const makeUrl = (params: UrlParams): string => {
-	return `${params.baseUrl}/${params.model}`;
+	return `${params.baseUrl}/v1/${params.model}`;
 };
 
 export const BLACK_FOREST_LABS_CONFIG: ProviderConfig = {

package/src/providers/cohere.ts

@@ -18,7 +18,6 @@ import type { ProviderConfig, UrlParams, HeaderParams, BodyParams } from "../typ
 
 const COHERE_API_BASE_URL = "https://api.cohere.com";
 
-
 const makeBody = (params: BodyParams): Record<string, unknown> => {
 	return {
 		...params.args,

package/src/providers/fireworks-ai.ts

@@ -16,7 +16,7 @@
  */
 import type { ProviderConfig, UrlParams, HeaderParams, BodyParams } from "../types";
 
-const FIREWORKS_AI_API_BASE_URL = "https://api.fireworks.ai/inference";
+const FIREWORKS_AI_API_BASE_URL = "https://api.fireworks.ai";
 
 const makeBody = (params: BodyParams): Record<string, unknown> => {
 	return {
@@ -30,10 +30,10 @@ const makeHeaders = (params: HeaderParams): Record<string, string> => {
 };
 
 const makeUrl = (params: UrlParams): string => {
-	if (params.task === "text-generation" && params.chatCompletion) {
-		return `${params.baseUrl}/v1/chat/completions`;
+	if (params.chatCompletion) {
+		return `${params.baseUrl}/inference/v1/chat/completions`;
 	}
-	return params.baseUrl;
+	return `${params.baseUrl}/inference`;
 };
 
 export const FIREWORKS_AI_CONFIG: ProviderConfig = {

package/src/providers/hf-inference.ts

@@ -29,7 +29,7 @@ const makeUrl = (params: UrlParams): string => {
 		/// when deployed on hf-inference, those two tasks are automatically compatible with one another.
 		return `${params.baseUrl}/pipeline/${params.task}/${params.model}`;
 	}
-	if (params.task === "text-generation" && params.chatCompletion) {
+	if (params.chatCompletion) {
 		return `${params.baseUrl}/models/${params.model}/v1/chat/completions`;
 	}
 	return `${params.baseUrl}/models/${params.model}`;

package/src/providers/nebius.ts

@@ -33,10 +33,10 @@ const makeUrl = (params: UrlParams): string => {
 	if (params.task === "text-to-image") {
 		return `${params.baseUrl}/v1/images/generations`;
 	}
+	if (params.chatCompletion) {
+		return `${params.baseUrl}/v1/chat/completions`;
+	}
 	if (params.task === "text-generation") {
-		if (params.chatCompletion) {
-			return `${params.baseUrl}/v1/chat/completions`;
-		}
 		return `${params.baseUrl}/v1/completions`;
 	}
 	return params.baseUrl;

package/src/providers/novita.ts

@@ -16,7 +16,7 @@
  */
 import type { ProviderConfig, UrlParams, HeaderParams, BodyParams } from "../types";
 
-const NOVITA_API_BASE_URL = "https://api.novita.ai/v3/openai";
+const NOVITA_API_BASE_URL = "https://api.novita.ai";
 
 const makeBody = (params: BodyParams): Record<string, unknown> => {
 	return {
@@ -30,11 +30,12 @@ const makeHeaders = (params: HeaderParams): Record<string, string> => {
 };
 
 const makeUrl = (params: UrlParams): string => {
-	if (params.task === "text-generation") {
-		if (params.chatCompletion) {
-			return `${params.baseUrl}/chat/completions`;
-		}
-		return `${params.baseUrl}/completions`;
+	if (params.chatCompletion) {
+		return `${params.baseUrl}/v3/openai/chat/completions`;
+	} else if (params.task === "text-generation") {
+		return `${params.baseUrl}/v3/openai/completions`;
+	} else if (params.task === "text-to-video") {
+		return `${params.baseUrl}/v3/hf/${params.model}`;
 	}
 	return params.baseUrl;
 };

package/src/providers/sambanova.ts

@@ -30,7 +30,7 @@ const makeHeaders = (params: HeaderParams): Record<string, string> => {
 };
 
 const makeUrl = (params: UrlParams): string => {
-	if (params.task === "text-generation" && params.chatCompletion) {
+	if (params.chatCompletion) {
 		return `${params.baseUrl}/v1/chat/completions`;
 	}
 	return params.baseUrl;

package/src/providers/together.ts

@@ -33,10 +33,10 @@ const makeUrl = (params: UrlParams): string => {
 	if (params.task === "text-to-image") {
 		return `${params.baseUrl}/v1/images/generations`;
 	}
+	if (params.chatCompletion) {
+		return `${params.baseUrl}/v1/chat/completions`;
+	}
 	if (params.task === "text-generation") {
-		if (params.chatCompletion) {
-			return `${params.baseUrl}/v1/chat/completions`;
-		}
 		return `${params.baseUrl}/v1/completions`;
 	}
 	return params.baseUrl;