@huggingface/inference 3.3.6 → 3.4.0

package/dist/index.js

@@ -45,32 +45,277 @@ __export(tasks_exports, {
 var HF_HUB_URL = "https://huggingface.co";
 var HF_ROUTER_URL = "https://router.huggingface.co";
 
+// src/providers/black-forest-labs.ts
+var BLACK_FOREST_LABS_AI_API_BASE_URL = "https://api.us1.bfl.ai/v1";
+var makeBody = (params) => {
+  return params.args;
+};
+var makeHeaders = (params) => {
+  if (params.authMethod === "provider-key") {
+    return { "X-Key": `${params.accessToken}` };
+  } else {
+    return { Authorization: `Bearer ${params.accessToken}` };
+  }
+};
+var makeUrl = (params) => {
+  return `${params.baseUrl}/${params.model}`;
+};
+var BLACK_FOREST_LABS_CONFIG = {
+  baseUrl: BLACK_FOREST_LABS_AI_API_BASE_URL,
+  makeBody,
+  makeHeaders,
+  makeUrl
+};
+
+// src/providers/cohere.ts
+var COHERE_API_BASE_URL = "https://api.cohere.com";
+var makeBody2 = (params) => {
+  return {
+    ...params.args,
+    model: params.model
+  };
+};
+var makeHeaders2 = (params) => {
+  return { Authorization: `Bearer ${params.accessToken}` };
+};
+var makeUrl2 = (params) => {
+  return `${params.baseUrl}/compatibility/v1/chat/completions`;
+};
+var COHERE_CONFIG = {
+  baseUrl: COHERE_API_BASE_URL,
+  makeBody: makeBody2,
+  makeHeaders: makeHeaders2,
+  makeUrl: makeUrl2
+};
+
 // src/providers/fal-ai.ts
 var FAL_AI_API_BASE_URL = "https://fal.run";
+var makeBody3 = (params) => {
+  return params.args;
+};
+var makeHeaders3 = (params) => {
+  return {
+    Authorization: params.authMethod === "provider-key" ? `Key ${params.accessToken}` : `Bearer ${params.accessToken}`
+  };
+};
+var makeUrl3 = (params) => {
+  return `${params.baseUrl}/${params.model}`;
+};
+var FAL_AI_CONFIG = {
+  baseUrl: FAL_AI_API_BASE_URL,
+  makeBody: makeBody3,
+  makeHeaders: makeHeaders3,
+  makeUrl: makeUrl3
+};
+
+// src/providers/fireworks-ai.ts
+var FIREWORKS_AI_API_BASE_URL = "https://api.fireworks.ai/inference";
+var makeBody4 = (params) => {
+  return {
+    ...params.args,
+    ...params.chatCompletion ? { model: params.model } : void 0
+  };
+};
+var makeHeaders4 = (params) => {
+  return { Authorization: `Bearer ${params.accessToken}` };
+};
+var makeUrl4 = (params) => {
+  if (params.task === "text-generation" && params.chatCompletion) {
+    return `${params.baseUrl}/v1/chat/completions`;
+  }
+  return params.baseUrl;
+};
+var FIREWORKS_AI_CONFIG = {
+  baseUrl: FIREWORKS_AI_API_BASE_URL,
+  makeBody: makeBody4,
+  makeHeaders: makeHeaders4,
+  makeUrl: makeUrl4
+};
+
+// src/providers/hf-inference.ts
+var makeBody5 = (params) => {
+  return {
+    ...params.args,
+    ...params.chatCompletion ? { model: params.model } : void 0
+  };
+};
+var makeHeaders5 = (params) => {
+  return { Authorization: `Bearer ${params.accessToken}` };
+};
+var makeUrl5 = (params) => {
+  if (params.task && ["feature-extraction", "sentence-similarity"].includes(params.task)) {
+    return `${params.baseUrl}/pipeline/${params.task}/${params.model}`;
+  }
+  if (params.task === "text-generation" && params.chatCompletion) {
+    return `${params.baseUrl}/models/${params.model}/v1/chat/completions`;
+  }
+  return `${params.baseUrl}/models/${params.model}`;
+};
+var HF_INFERENCE_CONFIG = {
+  baseUrl: `${HF_ROUTER_URL}/hf-inference`,
+  makeBody: makeBody5,
+  makeHeaders: makeHeaders5,
+  makeUrl: makeUrl5
+};
+
+// src/providers/hyperbolic.ts
+var HYPERBOLIC_API_BASE_URL = "https://api.hyperbolic.xyz";
+var makeBody6 = (params) => {
+  return {
+    ...params.args,
+    ...params.task === "text-to-image" ? { model_name: params.model } : { model: params.model }
+  };
+};
+var makeHeaders6 = (params) => {
+  return { Authorization: `Bearer ${params.accessToken}` };
+};
+var makeUrl6 = (params) => {
+  if (params.task === "text-to-image") {
+    return `${params.baseUrl}/v1/images/generations`;
+  }
+  return `${params.baseUrl}/v1/chat/completions`;
+};
+var HYPERBOLIC_CONFIG = {
+  baseUrl: HYPERBOLIC_API_BASE_URL,
+  makeBody: makeBody6,
+  makeHeaders: makeHeaders6,
+  makeUrl: makeUrl6
+};
 
 // src/providers/nebius.ts
 var NEBIUS_API_BASE_URL = "https://api.studio.nebius.ai";
+var makeBody7 = (params) => {
+  return {
+    ...params.args,
+    model: params.model
+  };
+};
+var makeHeaders7 = (params) => {
+  return { Authorization: `Bearer ${params.accessToken}` };
+};
+var makeUrl7 = (params) => {
+  if (params.task === "text-to-image") {
+    return `${params.baseUrl}/v1/images/generations`;
+  }
+  if (params.task === "text-generation") {
+    if (params.chatCompletion) {
+      return `${params.baseUrl}/v1/chat/completions`;
+    }
+    return `${params.baseUrl}/v1/completions`;
+  }
+  return params.baseUrl;
+};
+var NEBIUS_CONFIG = {
+  baseUrl: NEBIUS_API_BASE_URL,
+  makeBody: makeBody7,
+  makeHeaders: makeHeaders7,
+  makeUrl: makeUrl7
+};
+
+// src/providers/novita.ts
+var NOVITA_API_BASE_URL = "https://api.novita.ai/v3/openai";
+var makeBody8 = (params) => {
+  return {
+    ...params.args,
+    ...params.chatCompletion ? { model: params.model } : void 0
+  };
+};
+var makeHeaders8 = (params) => {
+  return { Authorization: `Bearer ${params.accessToken}` };
+};
+var makeUrl8 = (params) => {
+  if (params.task === "text-generation") {
+    if (params.chatCompletion) {
+      return `${params.baseUrl}/chat/completions`;
+    }
+    return `${params.baseUrl}/completions`;
+  }
+  return params.baseUrl;
+};
+var NOVITA_CONFIG = {
+  baseUrl: NOVITA_API_BASE_URL,
+  makeBody: makeBody8,
+  makeHeaders: makeHeaders8,
+  makeUrl: makeUrl8
+};
 
 // src/providers/replicate.ts
 var REPLICATE_API_BASE_URL = "https://api.replicate.com";
+var makeBody9 = (params) => {
+  return {
+    input: params.args,
+    version: params.model.includes(":") ? params.model.split(":")[1] : void 0
+  };
+};
+var makeHeaders9 = (params) => {
+  return { Authorization: `Bearer ${params.accessToken}` };
+};
+var makeUrl9 = (params) => {
+  if (params.model.includes(":")) {
+    return `${params.baseUrl}/v1/predictions`;
+  }
+  return `${params.baseUrl}/v1/models/${params.model}/predictions`;
+};
+var REPLICATE_CONFIG = {
+  baseUrl: REPLICATE_API_BASE_URL,
+  makeBody: makeBody9,
+  makeHeaders: makeHeaders9,
+  makeUrl: makeUrl9
+};
 
 // src/providers/sambanova.ts
 var SAMBANOVA_API_BASE_URL = "https://api.sambanova.ai";
+var makeBody10 = (params) => {
+  return {
+    ...params.args,
+    ...params.chatCompletion ? { model: params.model } : void 0
+  };
+};
+var makeHeaders10 = (params) => {
+  return { Authorization: `Bearer ${params.accessToken}` };
+};
+var makeUrl10 = (params) => {
+  if (params.task === "text-generation" && params.chatCompletion) {
+    return `${params.baseUrl}/v1/chat/completions`;
+  }
+  return params.baseUrl;
+};
+var SAMBANOVA_CONFIG = {
+  baseUrl: SAMBANOVA_API_BASE_URL,
+  makeBody: makeBody10,
+  makeHeaders: makeHeaders10,
+  makeUrl: makeUrl10
+};
 
 // src/providers/together.ts
 var TOGETHER_API_BASE_URL = "https://api.together.xyz";
-
-// src/providers/novita.ts
-var NOVITA_API_BASE_URL = "https://api.novita.ai/v3/openai";
-
-// src/providers/fireworks-ai.ts
-var FIREWORKS_AI_API_BASE_URL = "https://api.fireworks.ai/inference";
-
-// src/providers/hyperbolic.ts
-var HYPERBOLIC_API_BASE_URL = "https://api.hyperbolic.xyz";
-
-// src/providers/black-forest-labs.ts
-var BLACKFORESTLABS_AI_API_BASE_URL = "https://api.us1.bfl.ai/v1";
+var makeBody11 = (params) => {
+  return {
+    ...params.args,
+    model: params.model
+  };
+};
+var makeHeaders11 = (params) => {
+  return { Authorization: `Bearer ${params.accessToken}` };
+};
+var makeUrl11 = (params) => {
+  if (params.task === "text-to-image") {
+    return `${params.baseUrl}/v1/images/generations`;
+  }
+  if (params.task === "text-generation") {
+    if (params.chatCompletion) {
+      return `${params.baseUrl}/v1/chat/completions`;
+    }
+    return `${params.baseUrl}/v1/completions`;
+  }
+  return params.baseUrl;
+};
+var TOGETHER_CONFIG = {
+  baseUrl: TOGETHER_API_BASE_URL,
+  makeBody: makeBody11,
+  makeHeaders: makeHeaders11,
+  makeUrl: makeUrl11
+};
 
 // src/lib/isUrl.ts
 function isUrl(modelOrUrl) {
@@ -79,7 +324,7 @@ function isUrl(modelOrUrl) {
 
 // package.json
 var name = "@huggingface/inference";
-var version = "3.3.6";
+var version = "3.4.0";
 
 // src/providers/consts.ts
 var HARDCODED_MODEL_ID_MAPPING = {
@@ -90,15 +335,16 @@ var HARDCODED_MODEL_ID_MAPPING = {
    * "Qwen/Qwen2.5-Coder-32B-Instruct": "Qwen2.5-Coder-32B-Instruct",
    */
   "black-forest-labs": {},
+  cohere: {},
   "fal-ai": {},
   "fireworks-ai": {},
   "hf-inference": {},
   hyperbolic: {},
   nebius: {},
+  novita: {},
   replicate: {},
   sambanova: {},
-  together: {},
-  novita: {}
+  together: {}
 };
 
 // src/lib/getProviderModelId.ts
@@ -107,10 +353,10 @@ async function getProviderModelId(params, args, options = {}) {
   if (params.provider === "hf-inference") {
     return params.model;
   }
-  if (!options.taskHint) {
-    throw new Error("taskHint must be specified when using a third-party provider");
+  if (!options.task) {
+    throw new Error("task must be specified when using a third-party provider");
   }
-  const task = options.taskHint === "text-generation" && options.chatCompletion ? "conversational" : options.taskHint;
+  const task = options.task === "text-generation" && options.chatCompletion ? "conversational" : options.task;
   if (HARDCODED_MODEL_ID_MAPPING[params.provider]?.[params.model]) {
     return HARDCODED_MODEL_ID_MAPPING[params.provider][params.model];
   }
@@ -148,165 +394,83 @@ async function getProviderModelId(params, args, options = {}) {
 // src/lib/makeRequestOptions.ts
 var HF_HUB_INFERENCE_PROXY_TEMPLATE = `${HF_ROUTER_URL}/{{PROVIDER}}`;
 var tasks = null;
+var providerConfigs = {
+  "black-forest-labs": BLACK_FOREST_LABS_CONFIG,
+  cohere: COHERE_CONFIG,
+  "fal-ai": FAL_AI_CONFIG,
+  "fireworks-ai": FIREWORKS_AI_CONFIG,
+  "hf-inference": HF_INFERENCE_CONFIG,
+  hyperbolic: HYPERBOLIC_CONFIG,
+  nebius: NEBIUS_CONFIG,
+  novita: NOVITA_CONFIG,
+  replicate: REPLICATE_CONFIG,
+  sambanova: SAMBANOVA_CONFIG,
+  together: TOGETHER_CONFIG
+};
 async function makeRequestOptions(args, options) {
   const { accessToken, endpointUrl, provider: maybeProvider, model: maybeModel, ...remainingArgs } = args;
-  let otherArgs = remainingArgs;
   const provider = maybeProvider ?? "hf-inference";
-  const { includeCredentials, taskHint, chatCompletion: chatCompletion2 } = options ?? {};
+  const providerConfig = providerConfigs[provider];
+  const { includeCredentials, task, chatCompletion: chatCompletion2, signal } = options ?? {};
   if (endpointUrl && provider !== "hf-inference") {
     throw new Error(`Cannot use endpointUrl with a third-party provider.`);
   }
   if (maybeModel && isUrl(maybeModel)) {
     throw new Error(`Model URLs are no longer supported. Use endpointUrl instead.`);
   }
-  if (!maybeModel && !taskHint) {
+  if (!maybeModel && !task) {
     throw new Error("No model provided, and no task has been specified.");
   }
-  const hfModel = maybeModel ?? await loadDefaultModel(taskHint);
+  if (!providerConfig) {
+    throw new Error(`No provider config found for provider ${provider}`);
+  }
+  const hfModel = maybeModel ?? await loadDefaultModel(task);
   const model = await getProviderModelId({ model: hfModel, provider }, args, {
-    taskHint,
+    task,
     chatCompletion: chatCompletion2,
     fetch: options?.fetch
   });
   const authMethod = accessToken ? accessToken.startsWith("hf_") ? "hf-token" : "provider-key" : includeCredentials === "include" ? "credentials-include" : "none";
-  const url = endpointUrl ? chatCompletion2 ? endpointUrl + `/v1/chat/completions` : endpointUrl : makeUrl({
-    authMethod,
-    chatCompletion: chatCompletion2 ?? false,
+  const url = endpointUrl ? chatCompletion2 ? endpointUrl + `/v1/chat/completions` : endpointUrl : providerConfig.makeUrl({
+    baseUrl: authMethod !== "provider-key" ? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", provider) : providerConfig.baseUrl,
     model,
-    provider: provider ?? "hf-inference",
-    taskHint
+    chatCompletion: chatCompletion2,
+    task
   });
-  const headers = {};
-  if (accessToken) {
-    if (provider === "fal-ai" && authMethod === "provider-key") {
-      headers["Authorization"] = `Key ${accessToken}`;
-    } else if (provider === "black-forest-labs" && authMethod === "provider-key") {
-      headers["X-Key"] = accessToken;
-    } else {
-      headers["Authorization"] = `Bearer ${accessToken}`;
-    }
-  }
-  const ownUserAgent = `${name}/${version}`;
-  headers["User-Agent"] = [ownUserAgent, typeof navigator !== "undefined" ? navigator.userAgent : void 0].filter((x) => x !== void 0).join(" ");
   const binary = "data" in args && !!args.data;
+  const headers = providerConfig.makeHeaders({
+    accessToken,
+    authMethod
+  });
   if (!binary) {
     headers["Content-Type"] = "application/json";
   }
-  if (provider === "replicate") {
-    headers["Prefer"] = "wait";
-  }
+  const ownUserAgent = `${name}/${version}`;
+  const userAgent = [ownUserAgent, typeof navigator !== "undefined" ? navigator.userAgent : void 0].filter((x) => x !== void 0).join(" ");
+  headers["User-Agent"] = userAgent;
+  const body = binary ? args.data : JSON.stringify(
+    providerConfig.makeBody({
+      args: remainingArgs,
+      model,
+      task,
+      chatCompletion: chatCompletion2
+    })
+  );
   let credentials;
   if (typeof includeCredentials === "string") {
     credentials = includeCredentials;
   } else if (includeCredentials === true) {
     credentials = "include";
   }
-  if (provider === "replicate") {
-    const version2 = model.includes(":") ? model.split(":")[1] : void 0;
-    otherArgs = { input: otherArgs, version: version2 };
-  }
   const info = {
     headers,
     method: "POST",
-    body: binary ? args.data : JSON.stringify({
-      ...otherArgs,
-      ...taskHint === "text-to-image" && provider === "hyperbolic" ? { model_name: model } : chatCompletion2 || provider === "together" || provider === "nebius" || provider === "hyperbolic" ? { model } : void 0
-    }),
+    body,
     ...credentials ? { credentials } : void 0,
-    signal: options?.signal
+    signal
   };
   return { url, info };
 }
-function makeUrl(params) {
-  if (params.authMethod === "none" && params.provider !== "hf-inference") {
-    throw new Error("Authentication is required when requesting a third-party provider. Please provide accessToken");
-  }
-  const shouldProxy = params.provider !== "hf-inference" && params.authMethod !== "provider-key";
-  switch (params.provider) {
-    case "black-forest-labs": {
-      const baseUrl = shouldProxy ? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", params.provider) : BLACKFORESTLABS_AI_API_BASE_URL;
-      return `${baseUrl}/${params.model}`;
-    }
-    case "fal-ai": {
-      const baseUrl = shouldProxy ? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", params.provider) : FAL_AI_API_BASE_URL;
-      return `${baseUrl}/${params.model}`;
-    }
-    case "nebius": {
-      const baseUrl = shouldProxy ? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", params.provider) : NEBIUS_API_BASE_URL;
-      if (params.taskHint === "text-to-image") {
-        return `${baseUrl}/v1/images/generations`;
-      }
-      if (params.taskHint === "text-generation") {
-        if (params.chatCompletion) {
-          return `${baseUrl}/v1/chat/completions`;
-        }
-        return `${baseUrl}/v1/completions`;
-      }
-      return baseUrl;
-    }
-    case "replicate": {
-      const baseUrl = shouldProxy ? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", params.provider) : REPLICATE_API_BASE_URL;
-      if (params.model.includes(":")) {
-        return `${baseUrl}/v1/predictions`;
-      }
-      return `${baseUrl}/v1/models/${params.model}/predictions`;
-    }
-    case "sambanova": {
-      const baseUrl = shouldProxy ? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", params.provider) : SAMBANOVA_API_BASE_URL;
-      if (params.taskHint === "text-generation" && params.chatCompletion) {
-        return `${baseUrl}/v1/chat/completions`;
-      }
-      return baseUrl;
-    }
-    case "together": {
-      const baseUrl = shouldProxy ? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", params.provider) : TOGETHER_API_BASE_URL;
-      if (params.taskHint === "text-to-image") {
-        return `${baseUrl}/v1/images/generations`;
-      }
-      if (params.taskHint === "text-generation") {
-        if (params.chatCompletion) {
-          return `${baseUrl}/v1/chat/completions`;
-        }
-        return `${baseUrl}/v1/completions`;
-      }
-      return baseUrl;
-    }
-    case "fireworks-ai": {
-      const baseUrl = shouldProxy ? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", params.provider) : FIREWORKS_AI_API_BASE_URL;
-      if (params.taskHint === "text-generation" && params.chatCompletion) {
-        return `${baseUrl}/v1/chat/completions`;
-      }
-      return baseUrl;
-    }
-    case "hyperbolic": {
-      const baseUrl = shouldProxy ? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", params.provider) : HYPERBOLIC_API_BASE_URL;
-      if (params.taskHint === "text-to-image") {
-        return `${baseUrl}/v1/images/generations`;
-      }
-      return `${baseUrl}/v1/chat/completions`;
-    }
-    case "novita": {
-      const baseUrl = shouldProxy ? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", params.provider) : NOVITA_API_BASE_URL;
-      if (params.taskHint === "text-generation") {
-        if (params.chatCompletion) {
-          return `${baseUrl}/chat/completions`;
-        }
-        return `${baseUrl}/completions`;
-      }
-      return baseUrl;
-    }
-    default: {
-      const baseUrl = HF_HUB_INFERENCE_PROXY_TEMPLATE.replaceAll("{{PROVIDER}}", "hf-inference");
-      if (params.taskHint && ["feature-extraction", "sentence-similarity"].includes(params.taskHint)) {
-        return `${baseUrl}/pipeline/${params.taskHint}/${params.model}`;
-      }
-      if (params.taskHint === "text-generation" && params.chatCompletion) {
-        return `${baseUrl}/models/${params.model}/v1/chat/completions`;
-      }
-      return `${baseUrl}/models/${params.model}`;
-    }
-  }
-}
 async function loadDefaultModel(task) {
   if (!tasks) {
     tasks = await loadTaskInfo();
@@ -573,7 +737,7 @@ async function audioClassification(args, options) {
   const payload = preparePayload(args);
   const res = await request(payload, {
     ...options,
-    taskHint: "audio-classification"
+    task: "audio-classification"
   });
   const isValidOutput = Array.isArray(res) && res.every((x) => typeof x.label === "string" && typeof x.score === "number");
   if (!isValidOutput) {
@@ -600,7 +764,7 @@ async function automaticSpeechRecognition(args, options) {
   const payload = await buildPayload(args);
   const res = await request(payload, {
     ...options,
-    taskHint: "automatic-speech-recognition"
+    task: "automatic-speech-recognition"
   });
   const isValidOutput = typeof res?.text === "string";
   if (!isValidOutput) {
@@ -644,7 +808,7 @@ async function textToSpeech(args, options) {
   } : args;
   const res = await request(payload, {
     ...options,
-    taskHint: "text-to-speech"
+    task: "text-to-speech"
   });
   if (res instanceof Blob) {
     return res;
@@ -670,7 +834,7 @@ async function audioToAudio(args, options) {
   const payload = preparePayload(args);
   const res = await request(payload, {
     ...options,
-    taskHint: "audio-to-audio"
+    task: "audio-to-audio"
   });
   return validateOutput(res);
 }
@@ -696,7 +860,7 @@ async function imageClassification(args, options) {
   const payload = preparePayload2(args);
   const res = await request(payload, {
     ...options,
-    taskHint: "image-classification"
+    task: "image-classification"
   });
   const isValidOutput = Array.isArray(res) && res.every((x) => typeof x.label === "string" && typeof x.score === "number");
   if (!isValidOutput) {
@@ -710,7 +874,7 @@ async function imageSegmentation(args, options) {
   const payload = preparePayload2(args);
   const res = await request(payload, {
     ...options,
-    taskHint: "image-segmentation"
+    task: "image-segmentation"
   });
   const isValidOutput = Array.isArray(res) && res.every((x) => typeof x.label === "string" && typeof x.mask === "string" && typeof x.score === "number");
   if (!isValidOutput) {
@@ -724,7 +888,7 @@ async function imageToText(args, options) {
   const payload = preparePayload2(args);
   const res = (await request(payload, {
     ...options,
-    taskHint: "image-to-text"
+    task: "image-to-text"
   }))?.[0];
   if (typeof res?.generated_text !== "string") {
     throw new InferenceOutputError("Expected {generated_text: string}");
@@ -737,7 +901,7 @@ async function objectDetection(args, options) {
   const payload = preparePayload2(args);
   const res = await request(payload, {
     ...options,
-    taskHint: "object-detection"
+    task: "object-detection"
   });
   const isValidOutput = Array.isArray(res) && res.every(
     (x) => typeof x.label === "string" && typeof x.score === "number" && typeof x.box.xmin === "number" && typeof x.box.ymin === "number" && typeof x.box.xmax === "number" && typeof x.box.ymax === "number"
@@ -781,7 +945,7 @@ async function textToImage(args, options) {
   };
   const res = await request(payload, {
     ...options,
-    taskHint: "text-to-image"
+    task: "text-to-image"
   });
   if (res && typeof res === "object") {
     if (args.provider === "black-forest-labs" && "polling_url" in res && typeof res.polling_url === "string") {
@@ -870,7 +1034,7 @@ async function imageToImage(args, options) {
   }
   const res = await request(reqArgs, {
     ...options,
-    taskHint: "image-to-image"
+    task: "image-to-image"
   });
   const isValidOutput = res && res instanceof Blob;
   if (!isValidOutput) {
@@ -905,7 +1069,7 @@ async function zeroShotImageClassification(args, options) {
   const payload = await preparePayload3(args);
   const res = await request(payload, {
     ...options,
-    taskHint: "zero-shot-image-classification"
+    task: "zero-shot-image-classification"
   });
   const isValidOutput = Array.isArray(res) && res.every((x) => typeof x.label === "string" && typeof x.score === "number");
   if (!isValidOutput) {
@@ -925,7 +1089,7 @@ async function textToVideo(args, options) {
   const payload = args.provider === "fal-ai" || args.provider === "replicate" ? { ...omit(args, ["inputs", "parameters"]), ...args.parameters, prompt: args.inputs } : args;
   const res = await request(payload, {
     ...options,
-    taskHint: "text-to-video"
+    task: "text-to-video"
   });
   if (args.provider === "fal-ai") {
     const isValidOutput = typeof res === "object" && !!res && "video" in res && typeof res.video === "object" && !!res.video && "url" in res.video && typeof res.video.url === "string" && isUrl(res.video.url);
@@ -948,7 +1112,7 @@ async function textToVideo(args, options) {
 async function featureExtraction(args, options) {
   const res = await request(args, {
     ...options,
-    taskHint: "feature-extraction"
+    task: "feature-extraction"
   });
   let isValidOutput = true;
   const isNumArrayRec = (arr, maxDepth, curDepth = 0) => {
@@ -971,7 +1135,7 @@ async function featureExtraction(args, options) {
 async function fillMask(args, options) {
   const res = await request(args, {
     ...options,
-    taskHint: "fill-mask"
+    task: "fill-mask"
   });
   const isValidOutput = Array.isArray(res) && res.every(
     (x) => typeof x.score === "number" && typeof x.sequence === "string" && typeof x.token === "number" && typeof x.token_str === "string"
@@ -988,7 +1152,7 @@ async function fillMask(args, options) {
 async function questionAnswering(args, options) {
   const res = await request(args, {
     ...options,
-    taskHint: "question-answering"
+    task: "question-answering"
   });
   const isValidOutput = Array.isArray(res) ? res.every(
     (elem) => typeof elem === "object" && !!elem && typeof elem.answer === "string" && typeof elem.end === "number" && typeof elem.score === "number" && typeof elem.start === "number"
@@ -1003,7 +1167,7 @@ async function questionAnswering(args, options) {
 async function sentenceSimilarity(args, options) {
   const res = await request(prepareInput(args), {
     ...options,
-    taskHint: "sentence-similarity"
+    task: "sentence-similarity"
   });
   const isValidOutput = Array.isArray(res) && res.every((x) => typeof x === "number");
   if (!isValidOutput) {
@@ -1023,7 +1187,7 @@ function prepareInput(args) {
 async function summarization(args, options) {
   const res = await request(args, {
     ...options,
-    taskHint: "summarization"
+    task: "summarization"
   });
   const isValidOutput = Array.isArray(res) && res.every((x) => typeof x?.summary_text === "string");
   if (!isValidOutput) {
@@ -1036,7 +1200,7 @@ async function summarization(args, options) {
 async function tableQuestionAnswering(args, options) {
   const res = await request(args, {
     ...options,
-    taskHint: "table-question-answering"
+    task: "table-question-answering"
   });
   const isValidOutput = Array.isArray(res) ? res.every((elem) => validate(elem)) : validate(res);
   if (!isValidOutput) {
@@ -1056,7 +1220,7 @@ function validate(elem) {
 async function textClassification(args, options) {
   const res = (await request(args, {
     ...options,
-    taskHint: "text-classification"
+    task: "text-classification"
   }))?.[0];
   const isValidOutput = Array.isArray(res) && res.every((x) => typeof x?.label === "string" && typeof x.score === "number");
   if (!isValidOutput) {
@@ -1079,7 +1243,7 @@ async function textGeneration(args, options) {
     args.prompt = args.inputs;
     const raw = await request(args, {
       ...options,
-      taskHint: "text-generation"
+      task: "text-generation"
     });
     const isValidOutput = typeof raw === "object" && "choices" in raw && Array.isArray(raw?.choices) && typeof raw?.model === "string";
     if (!isValidOutput) {
@@ -1100,7 +1264,7 @@ async function textGeneration(args, options) {
     };
     const raw = await request(payload, {
       ...options,
-      taskHint: "text-generation"
+      task: "text-generation"
     });
     const isValidOutput = typeof raw === "object" && "choices" in raw && Array.isArray(raw?.choices) && typeof raw?.model === "string";
     if (!isValidOutput) {
@@ -1114,7 +1278,7 @@ async function textGeneration(args, options) {
     const res = toArray(
       await request(args, {
         ...options,
-        taskHint: "text-generation"
+        task: "text-generation"
       })
     );
     const isValidOutput = Array.isArray(res) && res.every((x) => "generated_text" in x && typeof x?.generated_text === "string");
@@ -1129,7 +1293,7 @@ async function textGeneration(args, options) {
 async function* textGenerationStream(args, options) {
   yield* streamingRequest(args, {
     ...options,
-    taskHint: "text-generation"
+    task: "text-generation"
   });
 }
 
@@ -1138,7 +1302,7 @@ async function tokenClassification(args, options) {
   const res = toArray(
     await request(args, {
       ...options,
-      taskHint: "token-classification"
+      task: "token-classification"
     })
   );
   const isValidOutput = Array.isArray(res) && res.every(
@@ -1156,7 +1320,7 @@ async function tokenClassification(args, options) {
 async function translation(args, options) {
   const res = await request(args, {
     ...options,
-    taskHint: "translation"
+    task: "translation"
   });
   const isValidOutput = Array.isArray(res) && res.every((x) => typeof x?.translation_text === "string");
   if (!isValidOutput) {
@@ -1170,7 +1334,7 @@ async function zeroShotClassification(args, options) {
   const res = toArray(
     await request(args, {
       ...options,
-      taskHint: "zero-shot-classification"
+      task: "zero-shot-classification"
     })
   );
   const isValidOutput = Array.isArray(res) && res.every(
@@ -1186,7 +1350,7 @@ async function zeroShotClassification(args, options) {
 async function chatCompletion(args, options) {
   const res = await request(args, {
     ...options,
-    taskHint: "text-generation",
+    task: "text-generation",
     chatCompletion: true
   });
   const isValidOutput = typeof res === "object" && Array.isArray(res?.choices) && typeof res?.created === "number" && typeof res?.id === "string" && typeof res?.model === "string" && /// Together.ai and Nebius do not output a system_fingerprint
@@ -1201,7 +1365,7 @@ async function chatCompletion(args, options) {
 async function* chatCompletionStream(args, options) {
   yield* streamingRequest(args, {
     ...options,
-    taskHint: "text-generation",
+    task: "text-generation",
     chatCompletion: true
   });
 }
@@ -1219,7 +1383,7 @@ async function documentQuestionAnswering(args, options) {
   const res = toArray(
     await request(reqArgs, {
       ...options,
-      taskHint: "document-question-answering"
+      task: "document-question-answering"
     })
   );
   const isValidOutput = Array.isArray(res) && res.every(
@@ -1243,7 +1407,7 @@ async function visualQuestionAnswering(args, options) {
   };
   const res = await request(reqArgs, {
     ...options,
-    taskHint: "visual-question-answering"
+    task: "visual-question-answering"
   });
   const isValidOutput = Array.isArray(res) && res.every(
     (elem) => typeof elem === "object" && !!elem && typeof elem?.answer === "string" && typeof elem.score === "number"
@@ -1258,7 +1422,7 @@ async function visualQuestionAnswering(args, options) {
 async function tabularRegression(args, options) {
   const res = await request(args, {
     ...options,
-    taskHint: "tabular-regression"
+    task: "tabular-regression"
   });
   const isValidOutput = Array.isArray(res) && res.every((x) => typeof x === "number");
   if (!isValidOutput) {
@@ -1271,7 +1435,7 @@ async function tabularRegression(args, options) {
 async function tabularClassification(args, options) {
   const res = await request(args, {
     ...options,
-    taskHint: "tabular-classification"
+    task: "tabular-classification"
   });
   const isValidOutput = Array.isArray(res) && res.every((x) => typeof x === "number");
   if (!isValidOutput) {
@@ -1323,6 +1487,7 @@ var HfInferenceEndpoint = class {
 // src/types.ts
 var INFERENCE_PROVIDERS = [
   "black-forest-labs",
+  "cohere",
   "fal-ai",
   "fireworks-ai",
   "hf-inference",