@huggingface/inference - npm Package Compare versions

Comparing version 3.3.6 to 3.3.7

dist/index.js

@@ -48,29 +48,253 @@ var __defProp = Object.defineProperty;
 
+// 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/fal-ai.ts
 var FAL_AI_API_BASE_URL = "https://fal.run";
+var makeBody2 = (params) => {
+  return params.args;
+};
+var makeHeaders2 = (params) => {
+  return {
+    Authorization: params.authMethod === "provider-key" ? `Key ${params.accessToken}` : `Bearer ${params.accessToken}`
+  };
+};
+var makeUrl2 = (params) => {
+  return `${params.baseUrl}/${params.model}`;
+};
+var FAL_AI_CONFIG = {
+  baseUrl: FAL_AI_API_BASE_URL,
+  makeBody: makeBody2,
+  makeHeaders: makeHeaders2,
+  makeUrl: makeUrl2
+};
 
+// src/providers/fireworks-ai.ts
+var FIREWORKS_AI_API_BASE_URL = "https://api.fireworks.ai/inference";
+var makeBody3 = (params) => {
+  return {
+    ...params.args,
+    ...params.chatCompletion ? { model: params.model } : void 0
+  };
+};
+var makeHeaders3 = (params) => {
+  return { Authorization: `Bearer ${params.accessToken}` };
+};
+var makeUrl3 = (params) => {
+  if (params.task === "text-generation" && params.chatCompletion) {
+    return `${params.baseUrl}/v1/chat/completions`;
+  }
+  return params.baseUrl;
+};
+var FIREWORKS_AI_CONFIG = {
+  baseUrl: FIREWORKS_AI_API_BASE_URL,
+  makeBody: makeBody3,
+  makeHeaders: makeHeaders3,
+  makeUrl: makeUrl3
+};
+
+// src/providers/hf-inference.ts
+var makeBody4 = (params) => {
+  return {
+    ...params.args,
+    ...params.chatCompletion ? { model: params.model } : void 0
+  };
+};
+var makeHeaders4 = (params) => {
+  return { Authorization: `Bearer ${params.accessToken}` };
+};
+var makeUrl4 = (params) => {
+  if (params.task && ["feature-extraction", "sentence-similarity"].includes(params.task)) {
+    return `${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: makeBody4,
+  makeHeaders: makeHeaders4,
+  makeUrl: makeUrl4
+};
+
+// src/providers/hyperbolic.ts
+var HYPERBOLIC_API_BASE_URL = "https://api.hyperbolic.xyz";
+var makeBody5 = (params) => {
+  return {
+    ...params.args,
+    ...params.task === "text-to-image" ? { model_name: params.model } : { model: params.model }
+  };
+};
+var makeHeaders5 = (params) => {
+  return { Authorization: `Bearer ${params.accessToken}` };
+};
+var makeUrl5 = (params) => {
+  if (params.task === "text-to-image") {
+    return `${params.baseUrl}/v1/images/generations`;
+  }
+  return `${params.baseUrl}/v1/chat/completions`;
+};
+var HYPERBOLIC_CONFIG = {
+  baseUrl: HYPERBOLIC_API_BASE_URL,
+  makeBody: makeBody5,
+  makeHeaders: makeHeaders5,
+  makeUrl: makeUrl5
+};
+
 // src/providers/nebius.ts
 var NEBIUS_API_BASE_URL = "https://api.studio.nebius.ai";
+var makeBody6 = (params) => {
+  return {
+    ...params.args,
+    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`;
+  }
+  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: makeBody6,
+  makeHeaders: makeHeaders6,
+  makeUrl: makeUrl6
+};
 
+// src/providers/novita.ts
+var NOVITA_API_BASE_URL = "https://api.novita.ai/v3/openai";
+var makeBody7 = (params) => {
+  return {
+    ...params.args,
+    ...params.chatCompletion ? { model: params.model } : void 0
+  };
+};
+var makeHeaders7 = (params) => {
+  return { Authorization: `Bearer ${params.accessToken}` };
+};
+var makeUrl7 = (params) => {
+  if (params.task === "text-generation") {
+    if (params.chatCompletion) {
+      return `${params.baseUrl}/chat/completions`;
+    }
+    return `${params.baseUrl}/completions`;
+  }
+  return params.baseUrl;
+};
+var NOVITA_CONFIG = {
+  baseUrl: NOVITA_API_BASE_URL,
+  makeBody: makeBody7,
+  makeHeaders: makeHeaders7,
+  makeUrl: makeUrl7
+};
+
 // src/providers/replicate.ts
 var REPLICATE_API_BASE_URL = "https://api.replicate.com";
+var makeBody8 = (params) => {
+  return {
+    input: params.args,
+    version: params.model.includes(":") ? params.model.split(":")[1] : void 0
+  };
+};
+var makeHeaders8 = (params) => {
+  return { Authorization: `Bearer ${params.accessToken}` };
+};
+var makeUrl8 = (params) => {
+  if (params.model.includes(":")) {
+    return `${params.baseUrl}/v1/predictions`;
+  }
+  return `${params.baseUrl}/v1/models/${params.model}/predictions`;
+};
+var REPLICATE_CONFIG = {
+  baseUrl: REPLICATE_API_BASE_URL,
+  makeBody: makeBody8,
+  makeHeaders: makeHeaders8,
+  makeUrl: makeUrl8
+};
 
 // src/providers/sambanova.ts
 var SAMBANOVA_API_BASE_URL = "https://api.sambanova.ai";
+var makeBody9 = (params) => {
+  return {
+    ...params.args,
+    ...params.chatCompletion ? { model: params.model } : void 0
+  };
+};
+var makeHeaders9 = (params) => {
+  return { Authorization: `Bearer ${params.accessToken}` };
+};
+var makeUrl9 = (params) => {
+  if (params.task === "text-generation" && params.chatCompletion) {
+    return `${params.baseUrl}/v1/chat/completions`;
+  }
+  return params.baseUrl;
+};
+var SAMBANOVA_CONFIG = {
+  baseUrl: SAMBANOVA_API_BASE_URL,
+  makeBody: makeBody9,
+  makeHeaders: makeHeaders9,
+  makeUrl: makeUrl9
+};
 
 // src/providers/together.ts
 var TOGETHER_API_BASE_URL = "https://api.together.xyz";
+var makeBody10 = (params) => {
+  return {
+    ...params.args,
+    model: params.model
+  };
+};
+var makeHeaders10 = (params) => {
+  return { Authorization: `Bearer ${params.accessToken}` };
+};
+var makeUrl10 = (params) => {
+  if (params.task === "text-to-image") {
+    return `${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: makeBody10,
+  makeHeaders: makeHeaders10,
+  makeUrl: makeUrl10
+};
 
-// 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";
-
 // src/lib/isUrl.ts
@@ -83,3 +307,3 @@ function isUrl(modelOrUrl) {
 var name = "@huggingface/inference";
-var version = "3.3.6";
+var version = "3.3.7";
 
@@ -100,6 +324,6 @@ // src/providers/consts.ts
   nebius: {},
+  novita: {},
   replicate: {},
   sambanova: {},
-  together: {},
-  novita: {}
+  together: {}
 };
@@ -113,6 +337,6 @@
   }
-  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]) {
@@ -155,7 +379,19 @@ return HARDCODED_MODEL_ID_MAPPING[params.provider][params.model];
 var tasks = null;
+var providerConfigs = {
+  "black-forest-labs": BLACK_FOREST_LABS_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") {
@@ -167,8 +403,11 @@ throw new Error(`Cannot use endpointUrl with a third-party provider.`);
   }
-  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,
@@ -178,28 +417,27 @@ 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}/${version}`;
+  const userAgent = [ownUserAgent, typeof navigator !== "undefined" ? navigator.userAgent : void 0].filter((x) => x !== void 0).join(" ");
+  headers["User-Agent"] = userAgent;
+  const body = binary ? args.data : JSON.stringify(
+    providerConfig.makeBody({
+      args: remainingArgs,
+      model,
+      task,
+      chatCompletion: chatCompletion2
+    })
+  );
   let credentials;
@@ -211,108 +449,11 @@ if (typeof includeCredentials === "string") {
   }
-  if (provider === "replicate") {
-    const version2 = model.includes(":") ? model.split(":")[1] : void 0;
-    otherArgs = { input: otherArgs, version: version2 };
-  }
   const info = {
     headers,
     method: "POST",
-    body: binary ? args.data : JSON.stringify({
-      ...otherArgs,
-      ...taskHint === "text-to-image" && provider === "hyperbolic" ? { model_name: model } : chatCompletion2 || provider === "together" || provider === "nebius" || provider === "hyperbolic" ? { model } : void 0
-    }),
+    body,
     ...credentials ? { credentials } : void 0,
-    signal: options?.signal
+    signal
   };
   return { url, info };
 }
-function makeUrl(params) {
-  if (params.authMethod === "none" && params.provider !== "hf-inference") {
-    throw new Error("Authentication is required when requesting a third-party provider. Please provide accessToken");
-  }
-  const shouldProxy = params.provider !== "hf-inference" && params.authMethod !== "provider-key";
-  switch (params.provider) {
-    case "black-forest-labs": {
-      const baseUrl = shouldProxy ? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", params.provider) : BLACKFORESTLABS_AI_API_BASE_URL;
-      return `${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) {
@@ -584,3 +725,3 @@ if (!tasks) {
     ...options,
-    taskHint: "audio-classification"
+    task: "audio-classification"
   });
@@ -612,3 +753,3 @@ const isValidOutput = Array.isArray(res) && res.every((x) => typeof x.label === "string" && typeof x.score === "number");
     ...options,
-    taskHint: "automatic-speech-recognition"
+    task: "automatic-speech-recognition"
   });
@@ -657,3 +798,3 @@ const isValidOutput = typeof res?.text === "string";
     ...options,
-    taskHint: "text-to-speech"
+    task: "text-to-speech"
   });
@@ -684,3 +825,3 @@ if (res instanceof Blob) {
     ...options,
-    taskHint: "audio-to-audio"
+    task: "audio-to-audio"
   });
@@ -711,3 +852,3 @@ return validateOutput(res);
     ...options,
-    taskHint: "image-classification"
+    task: "image-classification"
   });
@@ -726,3 +867,3 @@ const isValidOutput = Array.isArray(res) && res.every((x) => typeof x.label === "string" && typeof x.score === "number");
     ...options,
-    taskHint: "image-segmentation"
+    task: "image-segmentation"
   });
@@ -741,3 +882,3 @@ const isValidOutput = Array.isArray(res) && res.every((x) => typeof x.label === "string" && typeof x.mask === "string" && typeof x.score === "number");
     ...options,
-    taskHint: "image-to-text"
+    task: "image-to-text"
   }))?.[0];
@@ -755,3 +896,3 @@ if (typeof res?.generated_text !== "string") {
     ...options,
-    taskHint: "object-detection"
+    task: "object-detection"
   });
@@ -800,3 +941,3 @@ const isValidOutput = Array.isArray(res) && res.every(
     ...options,
-    taskHint: "text-to-image"
+    task: "text-to-image"
   });
@@ -890,3 +1031,3 @@ if (res && typeof res === "object") {
     ...options,
-    taskHint: "image-to-image"
+    task: "image-to-image"
   });
@@ -926,3 +1067,3 @@ const isValidOutput = res && res instanceof Blob;
     ...options,
-    taskHint: "zero-shot-image-classification"
+    task: "zero-shot-image-classification"
   });
@@ -947,3 +1088,3 @@ const isValidOutput = Array.isArray(res) && res.every((x) => typeof x.label === "string" && typeof x.score === "number");
     ...options,
-    taskHint: "text-to-video"
+    task: "text-to-video"
   });
@@ -971,3 +1112,3 @@ if (args.provider === "fal-ai") {
     ...options,
-    taskHint: "feature-extraction"
+    task: "feature-extraction"
   });
@@ -995,3 +1136,3 @@ let isValidOutput = true;
     ...options,
-    taskHint: "fill-mask"
+    task: "fill-mask"
   });
@@ -1013,3 +1154,3 @@ const isValidOutput = Array.isArray(res) && res.every(
     ...options,
-    taskHint: "question-answering"
+    task: "question-answering"
   });
@@ -1029,3 +1170,3 @@ const isValidOutput = Array.isArray(res) ? res.every(
     ...options,
-    taskHint: "sentence-similarity"
+    task: "sentence-similarity"
   });
@@ -1050,3 +1191,3 @@ const isValidOutput = Array.isArray(res) && res.every((x) => typeof x === "number");
     ...options,
-    taskHint: "summarization"
+    task: "summarization"
   });
@@ -1064,3 +1205,3 @@ const isValidOutput = Array.isArray(res) && res.every((x) => typeof x?.summary_text === "string");
     ...options,
-    taskHint: "table-question-answering"
+    task: "table-question-answering"
   });
@@ -1085,3 +1226,3 @@ const isValidOutput = Array.isArray(res) ? res.every((elem) => validate(elem)) : validate(res);
     ...options,
-    taskHint: "text-classification"
+    task: "text-classification"
   }))?.[0];
@@ -1109,3 +1250,3 @@ const isValidOutput = Array.isArray(res) && res.every((x) => typeof x?.label === "string" && typeof x.score === "number");
       ...options,
-      taskHint: "text-generation"
+      task: "text-generation"
     });
@@ -1131,3 +1272,3 @@ const isValidOutput = typeof raw === "object" && "choices" in raw && Array.isArray(raw?.choices) && typeof raw?.model === "string";
       ...options,
-      taskHint: "text-generation"
+      task: "text-generation"
     });
@@ -1146,3 +1287,3 @@ const isValidOutput = typeof raw === "object" && "choices" in raw && Array.isArray(raw?.choices) && typeof raw?.model === "string";
         ...options,
-        taskHint: "text-generation"
+        task: "text-generation"
       })
@@ -1162,3 +1303,3 @@ );
     ...options,
-    taskHint: "text-generation"
+    task: "text-generation"
   });
@@ -1172,3 +1313,3 @@ }
       ...options,
-      taskHint: "token-classification"
+      task: "token-classification"
     })
@@ -1191,3 +1332,3 @@ );
     ...options,
-    taskHint: "translation"
+    task: "translation"
   });
@@ -1206,3 +1347,3 @@ const isValidOutput = Array.isArray(res) && res.every((x) => typeof x?.translation_text === "string");
       ...options,
-      taskHint: "zero-shot-classification"
+      task: "zero-shot-classification"
     })
@@ -1223,3 +1364,3 @@ );
     ...options,
-    taskHint: "text-generation",
+    task: "text-generation",
     chatCompletion: true
@@ -1239,3 +1380,3 @@ });
     ...options,
-    taskHint: "text-generation",
+    task: "text-generation",
     chatCompletion: true
@@ -1258,3 +1399,3 @@ });
       ...options,
-      taskHint: "document-question-answering"
+      task: "document-question-answering"
     })
@@ -1283,3 +1424,3 @@ );
     ...options,
-    taskHint: "visual-question-answering"
+    task: "visual-question-answering"
   });
@@ -1299,3 +1440,3 @@ const isValidOutput = Array.isArray(res) && res.every(
     ...options,
-    taskHint: "tabular-regression"
+    task: "tabular-regression"
   });
@@ -1313,3 +1454,3 @@ const isValidOutput = Array.isArray(res) && res.every((x) => typeof x === "number");
     ...options,
-    taskHint: "tabular-classification"
+    task: "tabular-classification"
   });
@@ -1316,0 +1457,0 @@ const isValidOutput = Array.isArray(res) && res.every((x) => typeof x === "number");

dist/src/lib/getProviderModelId.d.ts

@@ -6,3 +6,3 @@ import type { InferenceProvider, InferenceTask, Options, RequestArgs } from "../types";
 }, args: RequestArgs, options?: {
-    taskHint?: InferenceTask;
+    task?: InferenceTask;
     chatCompletion?: boolean;
@@ -9,0 +9,0 @@ fetch?: Options["fetch"];

dist/src/lib/makeRequestOptions.d.ts

@@ -9,4 +9,4 @@ import type { InferenceTask, Options, RequestArgs } from "../types";
 }, options?: Options & {
-    /** To load default model if needed */
-    taskHint?: InferenceTask;
+    /** In most cases (unless we pass a endpointUrl) we know the task */
+    task?: InferenceTask;
     chatCompletion?: boolean;
@@ -13,0 +13,0 @@ }): Promise<{

dist/src/providers/black-forest-labs.d.ts

@@ -1,2 +0,1 @@
-export declare const BLACKFORESTLABS_AI_API_BASE_URL = "https://api.us1.bfl.ai/v1";
 /**
@@ -18,2 +17,4 @@ * See the registered mapping of HF model ID => Black Forest Labs model ID here:
  */
+import type { ProviderConfig } from "../types";
+export declare const BLACK_FOREST_LABS_CONFIG: ProviderConfig;
 //# sourceMappingURL=black-forest-labs.d.ts.map

dist/src/providers/fal-ai.d.ts

@@ -1,2 +0,1 @@
-export declare const FAL_AI_API_BASE_URL = "https://fal.run";
 /**
@@ -18,2 +17,4 @@ * See the registered mapping of HF model ID => Fal model ID here:
  */
+import type { ProviderConfig } from "../types";
+export declare const FAL_AI_CONFIG: ProviderConfig;
 //# sourceMappingURL=fal-ai.d.ts.map

dist/src/providers/fireworks-ai.d.ts

@@ -1,2 +0,1 @@
-export declare const FIREWORKS_AI_API_BASE_URL = "https://api.fireworks.ai/inference";
 /**
@@ -18,2 +17,4 @@ * See the registered mapping of HF model ID => Fireworks model ID here:
  */
+import type { ProviderConfig } from "../types";
+export declare const FIREWORKS_AI_CONFIG: ProviderConfig;
 //# sourceMappingURL=fireworks-ai.d.ts.map

dist/src/providers/hyperbolic.d.ts

@@ -1,2 +0,1 @@
-export declare const HYPERBOLIC_API_BASE_URL = "https://api.hyperbolic.xyz";
 /**
@@ -18,2 +17,4 @@ * See the registered mapping of HF model ID => Hyperbolic model ID here:
  */
+import type { ProviderConfig } from "../types";
+export declare const HYPERBOLIC_CONFIG: ProviderConfig;
 //# sourceMappingURL=hyperbolic.d.ts.map

dist/src/providers/nebius.d.ts

@@ -1,2 +0,1 @@
-export declare const NEBIUS_API_BASE_URL = "https://api.studio.nebius.ai";
 /**
@@ -18,2 +17,4 @@ * See the registered mapping of HF model ID => Nebius model ID here:
  */
+import type { ProviderConfig } from "../types";
+export declare const NEBIUS_CONFIG: ProviderConfig;
 //# sourceMappingURL=nebius.d.ts.map

dist/src/providers/novita.d.ts

@@ -1,2 +0,1 @@
-export declare const NOVITA_API_BASE_URL = "https://api.novita.ai/v3/openai";
 /**
@@ -18,2 +17,4 @@ * See the registered mapping of HF model ID => Novita model ID here:
  */
+import type { ProviderConfig } from "../types";
+export declare const NOVITA_CONFIG: ProviderConfig;
 //# sourceMappingURL=novita.d.ts.map

dist/src/providers/replicate.d.ts

@@ -1,2 +0,1 @@
-export declare const REPLICATE_API_BASE_URL = "https://api.replicate.com";
 /**
@@ -18,2 +17,5 @@ * See the registered mapping of HF model ID => Replicate model ID here:
  */
+import type { ProviderConfig } from "../types";
+export declare const REPLICATE_API_BASE_URL = "https://api.replicate.com";
+export declare const REPLICATE_CONFIG: ProviderConfig;
 //# sourceMappingURL=replicate.d.ts.map

dist/src/providers/sambanova.d.ts

@@ -1,2 +0,1 @@
-export declare const SAMBANOVA_API_BASE_URL = "https://api.sambanova.ai";
 /**
@@ -18,2 +17,4 @@ * See the registered mapping of HF model ID => Sambanova model ID here:
  */
+import type { ProviderConfig } from "../types";
+export declare const SAMBANOVA_CONFIG: ProviderConfig;
 //# sourceMappingURL=sambanova.d.ts.map

dist/src/providers/together.d.ts

@@ -1,2 +0,1 @@
-export declare const TOGETHER_API_BASE_URL = "https://api.together.xyz";
 /**
@@ -18,2 +17,4 @@ * See the registered mapping of HF model ID => Together model ID here:
  */
+import type { ProviderConfig } from "../types";
+export declare const TOGETHER_CONFIG: ProviderConfig;
 //# sourceMappingURL=together.d.ts.map

dist/src/tasks/custom/request.d.ts

@@ -6,6 +6,4 @@ import type { InferenceTask, Options, RequestArgs } from "../../types";
 export declare function request<T>(args: RequestArgs, options?: Options & {
-    /** When a model can be used for multiple tasks, and we want to run a non-default task */
-    task?: string | InferenceTask;
-    /** To load default model if needed */
-    taskHint?: InferenceTask;
+    /** In most cases (unless we pass a endpointUrl) we know the task */
+    task?: InferenceTask;
     /** Is chat completion compatible */
@@ -12,0 +10,0 @@ chatCompletion?: boolean;

dist/src/tasks/custom/streamingRequest.d.ts

@@ -6,6 +6,4 @@ import type { InferenceTask, Options, RequestArgs } from "../../types";
 export declare function streamingRequest<T>(args: RequestArgs, options?: Options & {
-    /** When a model can be used for multiple tasks, and we want to run a non-default task */
-    task?: string | InferenceTask;
-    /** To load default model if needed */
-    taskHint?: InferenceTask;
+    /** In most cases (unless we pass a endpointUrl) we know the task */
+    task?: InferenceTask;
     /** Is chat completion compatible */
@@ -12,0 +10,0 @@ chatCompletion?: boolean;

dist/src/tasks/nlp/featureExtraction.d.ts

@@ -0,11 +1,4 @@
+import type { FeatureExtractionInput } from "@huggingface/tasks";
 import type { BaseArgs, Options } from "../../types";
-export type FeatureExtractionArgs = BaseArgs & {
-    /**
-     *  The inputs is a string or a list of strings to get the features from.
-     *
-     *  inputs: "That is a happy person",
-     *
-     */
-    inputs: string | string[];
-};
+export type FeatureExtractionArgs = BaseArgs & FeatureExtractionInput;
 /**
@@ -12,0 +5,0 @@ * Returned values are a multidimensional array of floats (dimension depending on if you sent a string or a list of string, and if the automatic reduction, usually mean_pooling for instance was applied for you or not. This should be explained on the model's README).

dist/src/types.d.ts

@@ -1,2 +0,2 @@
-import type { ChatCompletionInput, PipelineType } from "@huggingface/tasks";
+import type { ChatCompletionInput, FeatureExtractionInput, PipelineType } from "@huggingface/tasks";
 /**
@@ -68,6 +68,27 @@ * HF model id, like "meta-llama/Llama-3.3-70B-Instruct"
     audio_url: string;
-} | ChatCompletionInput) & {
+} | ChatCompletionInput | FeatureExtractionInput) & {
     parameters?: Record<string, unknown>;
+};
+export interface ProviderConfig {
+    baseUrl: string;
+    makeBody: (params: BodyParams) => Record<string, unknown>;
+    makeHeaders: (params: HeaderParams) => Record<string, string>;
+    makeUrl: (params: UrlParams) => string;
+}
+export interface HeaderParams {
     accessToken?: string;
-};
+    authMethod: "none" | "hf-token" | "credentials-include" | "provider-key";
+}
+export interface UrlParams {
+    baseUrl: string;
+    model: string;
+    task?: InferenceTask;
+    chatCompletion?: boolean;
+}
+export interface BodyParams {
+    args: Record<string, unknown>;
+    chatCompletion?: boolean;
+    model: string;
+    task?: InferenceTask;
+}
 //# sourceMappingURL=types.d.ts.map

package.json

 {
   "name": "@huggingface/inference",
-  "version": "3.3.6",
+  "version": "3.3.7",
   "packageManager": "pnpm@8.10.5",
@@ -42,3 +42,3 @@ "license": "MIT",
   "dependencies": {
-    "@huggingface/tasks": "^0.16.0"
+    "@huggingface/tasks": "^0.16.3"
   },
@@ -45,0 +45,0 @@ "devDependencies": {

src/lib/getProviderModelId.ts

@@ -18,3 +18,3 @@ import type { WidgetType } from "@huggingface/tasks";
 	options: {
-		taskHint?: InferenceTask;
+		task?: InferenceTask;
 		chatCompletion?: boolean;
@@ -27,7 +27,7 @@ fetch?: Options["fetch"];
 	}
-	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: WidgetType =
-		options.taskHint === "text-generation" && options.chatCompletion ? "conversational" : options.taskHint;
+		options.task === "text-generation" && options.chatCompletion ? "conversational" : options.task;
 
@@ -34,0 +34,0 @@ // A dict called HARDCODED_MODEL_ID_MAPPING takes precedence in all cases (useful for dev purposes)

src/lib/makeRequestOptions.ts

 import { HF_HUB_URL, HF_ROUTER_URL } from "../config";
-import { FAL_AI_API_BASE_URL } from "../providers/fal-ai";
-import { NEBIUS_API_BASE_URL } from "../providers/nebius";
-import { REPLICATE_API_BASE_URL } from "../providers/replicate";
-import { SAMBANOVA_API_BASE_URL } from "../providers/sambanova";
-import { TOGETHER_API_BASE_URL } from "../providers/together";
-import { NOVITA_API_BASE_URL } from "../providers/novita";
-import { FIREWORKS_AI_API_BASE_URL } from "../providers/fireworks-ai";
-import { HYPERBOLIC_API_BASE_URL } from "../providers/hyperbolic";
-import { BLACKFORESTLABS_AI_API_BASE_URL } from "../providers/black-forest-labs";
-import type { InferenceProvider } from "../types";
-import type { InferenceTask, Options, RequestArgs } from "../types";
+import { BLACK_FOREST_LABS_CONFIG } from "../providers/black-forest-labs";
+import { FAL_AI_CONFIG } from "../providers/fal-ai";
+import { FIREWORKS_AI_CONFIG } from "../providers/fireworks-ai";
+import { HF_INFERENCE_CONFIG } from "../providers/hf-inference";
+import { HYPERBOLIC_CONFIG } from "../providers/hyperbolic";
+import { NEBIUS_CONFIG } from "../providers/nebius";
+import { NOVITA_CONFIG } from "../providers/novita";
+import { REPLICATE_CONFIG } from "../providers/replicate";
+import { SAMBANOVA_CONFIG } from "../providers/sambanova";
+import { TOGETHER_CONFIG } from "../providers/together";
+import type { InferenceProvider, InferenceTask, Options, ProviderConfig, RequestArgs } from "../types";
 import { isUrl } from "./isUrl";
@@ -26,2 +26,18 @@ import { version as packageVersion, name as packageName } from "../../package.json";
 /**
+ * Config to define how to serialize requests for each provider
+ */
+const providerConfigs: Record<InferenceProvider, ProviderConfig> = {
+	"black-forest-labs": BLACK_FOREST_LABS_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,
+};
+
+/**
  * Helper that prepares request arguments
@@ -35,4 +51,4 @@ */
 	options?: Options & {
-		/** To load default model if needed */
-		taskHint?: InferenceTask;
+		/** In most cases (unless we pass a endpointUrl) we know the task */
+		task?: InferenceTask;
 		chatCompletion?: boolean;
@@ -42,6 +58,6 @@ }
 	const { accessToken, endpointUrl, provider: maybeProvider, model: maybeModel, ...remainingArgs } = args;
-	let otherArgs = remainingArgs;
 	const provider = maybeProvider ?? "hf-inference";
+	const providerConfig = providerConfigs[provider];
 
-	const { includeCredentials, taskHint, chatCompletion } = options ?? {};
+	const { includeCredentials, task, chatCompletion, signal } = options ?? {};
 
@@ -54,9 +70,12 @@ if (endpointUrl && provider !== "hf-inference") {
 	}
-	if (!maybeModel && !taskHint) {
+	if (!maybeModel && !task) {
 		throw new Error("No model provided, and no task has been specified.");
 	}
+	if (!providerConfig) {
+		throw new Error(`No provider config found for provider ${provider}`);
+	}
 	// eslint-disable-next-line @typescript-eslint/no-non-null-assertion
-	const hfModel = maybeModel ?? (await loadDefaultModel(taskHint!));
+	const hfModel = maybeModel ?? (await loadDefaultModel(task!));
 	const model = await getProviderModelId({ model: hfModel, provider }, args, {
-		taskHint,
+		task,
 		chatCompletion,
@@ -75,2 +94,3 @@ fetch: options?.fetch,
 
+	// Make URL
 	const url = endpointUrl
@@ -80,37 +100,44 @@ ? chatCompletion
 			: endpointUrl
-		: makeUrl({
-				authMethod,
-				chatCompletion: chatCompletion ?? false,
+		: providerConfig.makeUrl({
+				baseUrl:
+					authMethod !== "provider-key"
+						? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", provider)
+						: providerConfig.baseUrl,
 				model,
-				provider: provider ?? "hf-inference",
-				taskHint,
+				chatCompletion,
+				task,
 		  });
 
-	const headers: Record<string, string> = {};
-	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}`;
-		}
+	// Make headers
+	const binary = "data" in args && !!args.data;
+	const headers = providerConfig.makeHeaders({
+		accessToken,
+		authMethod,
+	});
+
+	// Add content-type to headers
+	if (!binary) {
+		headers["Content-Type"] = "application/json";
 	}
 
+	// Add user-agent to headers
 	// e.g. @huggingface/inference/3.1.3
 	const ownUserAgent = `${packageName}/${packageVersion}`;
-	headers["User-Agent"] = [ownUserAgent, typeof navigator !== "undefined" ? navigator.userAgent : undefined]
+	const userAgent = [ownUserAgent, typeof navigator !== "undefined" ? navigator.userAgent : undefined]
 		.filter((x) => x !== undefined)
 		.join(" ");
+	headers["User-Agent"] = userAgent;
 
-	const binary = "data" in args && !!args.data;
+	// Make body
+	const body = binary
+		? args.data
+		: JSON.stringify(
+				providerConfig.makeBody({
+					args: remainingArgs as Record<string, unknown>,
+					model,
+					task,
+					chatCompletion,
+				})
+		  );
 
-	if (!binary) {
-		headers["Content-Type"] = "application/json";
-	}
-
-	if (provider === "replicate") {
-		headers["Prefer"] = "wait";
-	}
-
 	/**
@@ -126,26 +153,8 @@ * For edge runtimes, leave 'credentials' undefined, otherwise cloudflare workers will error
 
-	/**
-	 * Replicate models wrap all inputs inside { input: ... }
-	 * Versioned Replicate models in the format `owner/model:version` expect the version in the body
-	 */
-	if (provider === "replicate") {
-		const version = model.includes(":") ? model.split(":")[1] : undefined;
-		(otherArgs as unknown) = { input: otherArgs, version };
-	}
-
 	const info: RequestInit = {
 		headers,
 		method: "POST",
-		body: binary
-			? args.data
-			: JSON.stringify({
-					...otherArgs,
-					...(taskHint === "text-to-image" && provider === "hyperbolic"
-						? { model_name: model }
-						: chatCompletion || provider === "together" || provider === "nebius" || provider === "hyperbolic"
-						  ? { model }
-						  : undefined),
-			  }),
+		body,
 		...(credentials ? { credentials } : undefined),
-		signal: options?.signal,
+		signal,
 	};
@@ -156,125 +165,2 @@
 
-function makeUrl(params: {
-	authMethod: "none" | "hf-token" | "credentials-include" | "provider-key";
-	chatCompletion: boolean;
-	model: string;
-	provider: InferenceProvider;
-	taskHint: InferenceTask | undefined;
-}): string {
-	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 `${baseUrl}/v1/completions`;
-			}
-			return baseUrl;
-		}
-		case "replicate": {
-			const baseUrl = shouldProxy
-				? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", params.provider)
-				: REPLICATE_API_BASE_URL;
-			if (params.model.includes(":")) {
-				/// Versioned model
-				return `${baseUrl}/v1/predictions`;
-			}
-			/// Evergreen / Canonical model
-			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;
-			/// Sambanova API matches OpenAI-like APIs: model is defined in the request body
-			if (params.taskHint === "text-generation" && params.chatCompletion) {
-				return `${baseUrl}/v1/chat/completions`;
-			}
-			return baseUrl;
-		}
-		case "together": {
-			const baseUrl = shouldProxy
-				? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", params.provider)
-				: TOGETHER_API_BASE_URL;
-			/// Together API matches OpenAI-like APIs: model is defined in the request body
-			if (params.taskHint === "text-to-image") {
-				return `${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 `${baseUrl}/completions`;
-			}
-			return baseUrl;
-		}
-		default: {
-			const baseUrl = HF_HUB_INFERENCE_PROXY_TEMPLATE.replaceAll("{{PROVIDER}}", "hf-inference");
-			if (params.taskHint && ["feature-extraction", "sentence-similarity"].includes(params.taskHint)) {
-				/// when deployed on hf-inference, those two tasks are automatically compatible with one another.
-				return `${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: InferenceTask): Promise<string> {
@@ -281,0 +167,0 @@ if (!tasks) {

src/providers/black-forest-labs.ts

@@ -1,3 +0,1 @@
-export const BLACKFORESTLABS_AI_API_BASE_URL = "https://api.us1.bfl.ai/v1";
-
 /**
@@ -19,1 +17,27 @@ * See the registered mapping of HF model ID => Black Forest Labs model ID here:
  */
+import type { ProviderConfig, UrlParams, HeaderParams, BodyParams } from "../types";
+
+const BLACK_FOREST_LABS_AI_API_BASE_URL = "https://api.us1.bfl.ai/v1";
+
+const makeBody = (params: BodyParams): Record<string, unknown> => {
+	return params.args;
+};
+
+const makeHeaders = (params: HeaderParams): Record<string, string> => {
+	if (params.authMethod === "provider-key") {
+		return { "X-Key": `${params.accessToken}` };
+	} else {
+		return { Authorization: `Bearer ${params.accessToken}` };
+	}
+};
+
+const makeUrl = (params: UrlParams): string => {
+	return `${params.baseUrl}/${params.model}`;
+};
+
+export const BLACK_FOREST_LABS_CONFIG: ProviderConfig = {
+	baseUrl: BLACK_FOREST_LABS_AI_API_BASE_URL,
+	makeBody,
+	makeHeaders,
+	makeUrl,
+};

src/providers/consts.ts

@@ -25,6 +25,6 @@ import type { InferenceProvider } from "../types";
 	nebius: {},
+	novita: {},
 	replicate: {},
 	sambanova: {},
 	together: {},
-	novita: {},
 };

src/providers/fal-ai.ts

@@ -1,3 +0,1 @@
-export const FAL_AI_API_BASE_URL = "https://fal.run";
-
 /**
@@ -19,1 +17,25 @@ * See the registered mapping of HF model ID => Fal model ID here:
  */
+import type { ProviderConfig, UrlParams, HeaderParams, BodyParams } from "../types";
+
+const FAL_AI_API_BASE_URL = "https://fal.run";
+
+const makeBody = (params: BodyParams): Record<string, unknown> => {
+	return params.args;
+};
+
+const makeHeaders = (params: HeaderParams): Record<string, string> => {
+	return {
+		Authorization: params.authMethod === "provider-key" ? `Key ${params.accessToken}` : `Bearer ${params.accessToken}`,
+	};
+};
+
+const makeUrl = (params: UrlParams): string => {
+	return `${params.baseUrl}/${params.model}`;
+};
+
+export const FAL_AI_CONFIG: ProviderConfig = {
+	baseUrl: FAL_AI_API_BASE_URL,
+	makeBody,
+	makeHeaders,
+	makeUrl,
+};

src/providers/fireworks-ai.ts

@@ -1,3 +0,1 @@
-export const FIREWORKS_AI_API_BASE_URL = "https://api.fireworks.ai/inference";
-
 /**
@@ -19,1 +17,29 @@ * See the registered mapping of HF model ID => Fireworks model ID here:
  */
+import type { ProviderConfig, UrlParams, HeaderParams, BodyParams } from "../types";
+
+const FIREWORKS_AI_API_BASE_URL = "https://api.fireworks.ai/inference";
+
+const makeBody = (params: BodyParams): Record<string, unknown> => {
+	return {
+		...params.args,
+		...(params.chatCompletion ? { model: params.model } : undefined),
+	};
+};
+
+const makeHeaders = (params: HeaderParams): Record<string, string> => {
+	return { Authorization: `Bearer ${params.accessToken}` };
+};
+
+const makeUrl = (params: UrlParams): string => {
+	if (params.task === "text-generation" && params.chatCompletion) {
+		return `${params.baseUrl}/v1/chat/completions`;
+	}
+	return params.baseUrl;
+};
+
+export const FIREWORKS_AI_CONFIG: ProviderConfig = {
+	baseUrl: FIREWORKS_AI_API_BASE_URL,
+	makeBody,
+	makeHeaders,
+	makeUrl,
+};

src/providers/hyperbolic.ts

@@ -1,3 +0,1 @@
-export const HYPERBOLIC_API_BASE_URL = "https://api.hyperbolic.xyz";
-
 /**
@@ -19,1 +17,29 @@ * See the registered mapping of HF model ID => Hyperbolic model ID here:
  */
+import type { ProviderConfig, UrlParams, HeaderParams, BodyParams } from "../types";
+
+const HYPERBOLIC_API_BASE_URL = "https://api.hyperbolic.xyz";
+
+const makeBody = (params: BodyParams): Record<string, unknown> => {
+	return {
+		...params.args,
+		...(params.task === "text-to-image" ? { model_name: params.model } : { model: params.model }),
+	};
+};
+
+const makeHeaders = (params: HeaderParams): Record<string, string> => {
+	return { Authorization: `Bearer ${params.accessToken}` };
+};
+
+const makeUrl = (params: UrlParams): string => {
+	if (params.task === "text-to-image") {
+		return `${params.baseUrl}/v1/images/generations`;
+	}
+	return `${params.baseUrl}/v1/chat/completions`;
+};
+
+export const HYPERBOLIC_CONFIG: ProviderConfig = {
+	baseUrl: HYPERBOLIC_API_BASE_URL,
+	makeBody,
+	makeHeaders,
+	makeUrl,
+};

src/providers/nebius.ts

@@ -1,3 +0,1 @@
-export const NEBIUS_API_BASE_URL = "https://api.studio.nebius.ai";
-
 /**
@@ -19,1 +17,35 @@ * See the registered mapping of HF model ID => Nebius model ID here:
  */
+import type { ProviderConfig, UrlParams, HeaderParams, BodyParams } from "../types";
+
+const NEBIUS_API_BASE_URL = "https://api.studio.nebius.ai";
+
+const makeBody = (params: BodyParams): Record<string, unknown> => {
+	return {
+		...params.args,
+		model: params.model,
+	};
+};
+
+const makeHeaders = (params: HeaderParams): Record<string, string> => {
+	return { Authorization: `Bearer ${params.accessToken}` };
+};
+
+const makeUrl = (params: UrlParams): string => {
+	if (params.task === "text-to-image") {
+		return `${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;
+};
+
+export const NEBIUS_CONFIG: ProviderConfig = {
+	baseUrl: NEBIUS_API_BASE_URL,
+	makeBody,
+	makeHeaders,
+	makeUrl,
+};

src/providers/novita.ts

@@ -1,3 +0,1 @@
-export const NOVITA_API_BASE_URL = "https://api.novita.ai/v3/openai";
-
 /**
@@ -19,1 +17,32 @@ * See the registered mapping of HF model ID => Novita model ID here:
  */
+import type { ProviderConfig, UrlParams, HeaderParams, BodyParams } from "../types";
+
+const NOVITA_API_BASE_URL = "https://api.novita.ai/v3/openai";
+
+const makeBody = (params: BodyParams): Record<string, unknown> => {
+	return {
+		...params.args,
+		...(params.chatCompletion ? { model: params.model } : undefined),
+	};
+};
+
+const makeHeaders = (params: HeaderParams): Record<string, string> => {
+	return { Authorization: `Bearer ${params.accessToken}` };
+};
+
+const makeUrl = (params: UrlParams): string => {
+	if (params.task === "text-generation") {
+		if (params.chatCompletion) {
+			return `${params.baseUrl}/chat/completions`;
+		}
+		return `${params.baseUrl}/completions`;
+	}
+	return params.baseUrl;
+};
+
+export const NOVITA_CONFIG: ProviderConfig = {
+	baseUrl: NOVITA_API_BASE_URL,
+	makeBody,
+	makeHeaders,
+	makeUrl,
+};

src/providers/replicate.ts

@@ -1,3 +0,1 @@
-export const REPLICATE_API_BASE_URL = "https://api.replicate.com";
-
 /**
@@ -19,1 +17,31 @@ * See the registered mapping of HF model ID => Replicate model ID here:
  */
+import type { ProviderConfig, UrlParams, HeaderParams, BodyParams } from "../types";
+
+export const REPLICATE_API_BASE_URL = "https://api.replicate.com";
+
+const makeBody = (params: BodyParams): Record<string, unknown> => {
+	return {
+		input: params.args,
+		version: params.model.includes(":") ? params.model.split(":")[1] : undefined,
+	};
+};
+
+const makeHeaders = (params: HeaderParams): Record<string, string> => {
+	return { Authorization: `Bearer ${params.accessToken}` };
+};
+
+const makeUrl = (params: UrlParams): string => {
+	if (params.model.includes(":")) {
+		/// Versioned model
+		return `${params.baseUrl}/v1/predictions`;
+	}
+	/// Evergreen / Canonical model
+	return `${params.baseUrl}/v1/models/${params.model}/predictions`;
+};
+
+export const REPLICATE_CONFIG: ProviderConfig = {
+	baseUrl: REPLICATE_API_BASE_URL,
+	makeBody,
+	makeHeaders,
+	makeUrl,
+};

src/providers/sambanova.ts

@@ -1,3 +0,1 @@
-export const SAMBANOVA_API_BASE_URL = "https://api.sambanova.ai";
-
 /**
@@ -19,1 +17,29 @@ * See the registered mapping of HF model ID => Sambanova model ID here:
  */
+import type { ProviderConfig, UrlParams, HeaderParams, BodyParams } from "../types";
+
+const SAMBANOVA_API_BASE_URL = "https://api.sambanova.ai";
+
+const makeBody = (params: BodyParams): Record<string, unknown> => {
+	return {
+		...params.args,
+		...(params.chatCompletion ? { model: params.model } : undefined),
+	};
+};
+
+const makeHeaders = (params: HeaderParams): Record<string, string> => {
+	return { Authorization: `Bearer ${params.accessToken}` };
+};
+
+const makeUrl = (params: UrlParams): string => {
+	if (params.task === "text-generation" && params.chatCompletion) {
+		return `${params.baseUrl}/v1/chat/completions`;
+	}
+	return params.baseUrl;
+};
+
+export const SAMBANOVA_CONFIG: ProviderConfig = {
+	baseUrl: SAMBANOVA_API_BASE_URL,
+	makeBody,
+	makeHeaders,
+	makeUrl,
+};

src/providers/together.ts

@@ -1,3 +0,1 @@
-export const TOGETHER_API_BASE_URL = "https://api.together.xyz";
-
 /**
@@ -19,1 +17,35 @@ * See the registered mapping of HF model ID => Together model ID here:
  */
+import type { ProviderConfig, UrlParams, HeaderParams, BodyParams } from "../types";
+
+const TOGETHER_API_BASE_URL = "https://api.together.xyz";
+
+const makeBody = (params: BodyParams): Record<string, unknown> => {
+	return {
+		...params.args,
+		model: params.model,
+	};
+};
+
+const makeHeaders = (params: HeaderParams): Record<string, string> => {
+	return { Authorization: `Bearer ${params.accessToken}` };
+};
+
+const makeUrl = (params: UrlParams): string => {
+	if (params.task === "text-to-image") {
+		return `${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;
+};
+
+export const TOGETHER_CONFIG: ProviderConfig = {
+	baseUrl: TOGETHER_API_BASE_URL,
+	makeBody,
+	makeHeaders,
+	makeUrl,
+};

src/tasks/audio/audioClassification.ts

@@ -21,3 +21,3 @@ import type { AudioClassificationInput, AudioClassificationOutput } from "@huggingface/tasks";
 		...options,
-		taskHint: "audio-classification",
+		task: "audio-classification",
 	});
@@ -24,0 +24,0 @@ const isValidOutput =

src/tasks/audio/audioToAudio.ts

@@ -42,3 +42,3 @@ import { InferenceOutputError } from "../../lib/InferenceOutputError";
 		...options,
-		taskHint: "audio-to-audio",
+		task: "audio-to-audio",
 	});
@@ -45,0 +45,0 @@

src/tasks/audio/automaticSpeechRecognition.ts

@@ -22,3 +22,3 @@ import type { AutomaticSpeechRecognitionInput, AutomaticSpeechRecognitionOutput } from "@huggingface/tasks";
 		...options,
-		taskHint: "automatic-speech-recognition",
+		task: "automatic-speech-recognition",
 	});
@@ -25,0 +25,0 @@ const isValidOutput = typeof res?.text === "string";

src/tasks/audio/textToSpeech.ts

@@ -27,3 +27,3 @@ import type { TextToSpeechInput } from "@huggingface/tasks";
 		...options,
-		taskHint: "text-to-speech",
+		task: "text-to-speech",
 	});
@@ -30,0 +30,0 @@ if (res instanceof Blob) {

src/tasks/custom/request.ts

@@ -10,6 +10,4 @@ import type { InferenceTask, Options, RequestArgs } from "../../types";
 	options?: Options & {
-		/** When a model can be used for multiple tasks, and we want to run a non-default task */
-		task?: string | InferenceTask;
-		/** To load default model if needed */
-		taskHint?: InferenceTask;
+		/** In most cases (unless we pass a endpointUrl) we know the task */
+		task?: InferenceTask;
 		/** Is chat completion compatible */
@@ -16,0 +14,0 @@ chatCompletion?: boolean;

src/tasks/custom/streamingRequest.ts

@@ -12,6 +12,4 @@ import type { InferenceTask, Options, RequestArgs } from "../../types";
 	options?: Options & {
-		/** When a model can be used for multiple tasks, and we want to run a non-default task */
-		task?: string | InferenceTask;
-		/** To load default model if needed */
-		taskHint?: InferenceTask;
+		/** In most cases (unless we pass a endpointUrl) we know the task */
+		task?: InferenceTask;
 		/** Is chat completion compatible */
@@ -18,0 +16,0 @@ chatCompletion?: boolean;

src/tasks/cv/imageClassification.ts

@@ -20,3 +20,3 @@ import type { ImageClassificationInput, ImageClassificationOutput } from "@huggingface/tasks";
 		...options,
-		taskHint: "image-classification",
+		task: "image-classification",
 	});
@@ -23,0 +23,0 @@ const isValidOutput =

src/tasks/cv/imageSegmentation.ts

@@ -20,3 +20,3 @@ import type { ImageSegmentationInput, ImageSegmentationOutput } from "@huggingface/tasks";
 		...options,
-		taskHint: "image-segmentation",
+		task: "image-segmentation",
 	});
@@ -23,0 +23,0 @@ const isValidOutput =

src/tasks/cv/imageToImage.ts

@@ -31,3 +31,3 @@ import type { ImageToImageInput } from "@huggingface/tasks";
 		...options,
-		taskHint: "image-to-image",
+		task: "image-to-image",
 	});
@@ -34,0 +34,0 @@ const isValidOutput = res && res instanceof Blob;

src/tasks/cv/imageToText.ts

@@ -17,3 +17,3 @@ import type { ImageToTextInput, ImageToTextOutput } from "@huggingface/tasks";
 			...options,
-			taskHint: "image-to-text",
+			task: "image-to-text",
 		})
@@ -20,0 +20,0 @@ )?.[0];

src/tasks/cv/objectDetection.ts

@@ -17,3 +17,3 @@ import { request } from "../custom/request";
 		...options,
-		taskHint: "object-detection",
+		task: "object-detection",
 	});
@@ -20,0 +20,0 @@ const isValidOutput =

src/tasks/cv/textToImage.ts

@@ -76,3 +76,3 @@ import type { TextToImageInput, TextToImageOutput } from "@huggingface/tasks";
 		...options,
-		taskHint: "text-to-image",
+		task: "text-to-image",
 	});
@@ -79,0 +79,0 @@

src/tasks/cv/textToVideo.ts

@@ -38,3 +38,3 @@ import type { BaseArgs, InferenceProvider, Options } from "../../types";
 		...options,
-		taskHint: "text-to-video",
+		task: "text-to-video",
 	});
@@ -41,0 +41,0 @@

src/tasks/cv/zeroShotImageClassification.ts

@@ -51,3 +51,3 @@ import { InferenceOutputError } from "../../lib/InferenceOutputError";
 		...options,
-		taskHint: "zero-shot-image-classification",
+		task: "zero-shot-image-classification",
 	});
@@ -54,0 +54,0 @@ const isValidOutput =

src/tasks/multimodal/documentQuestionAnswering.ts

@@ -35,3 +35,3 @@ import { InferenceOutputError } from "../../lib/InferenceOutputError";
 			...options,
-			taskHint: "document-question-answering",
+			task: "document-question-answering",
 		})
@@ -38,0 +38,0 @@ );

src/tasks/multimodal/visualQuestionAnswering.ts

@@ -32,3 +32,3 @@ import type {
 		...options,
-		taskHint: "visual-question-answering",
+		task: "visual-question-answering",
 	});
@@ -35,0 +35,0 @@ const isValidOutput =

src/tasks/nlp/chatCompletion.ts

@@ -15,3 +15,3 @@ import { InferenceOutputError } from "../../lib/InferenceOutputError";
 		...options,
-		taskHint: "text-generation",
+		task: "text-generation",
 		chatCompletion: true,
@@ -18,0 +18,0 @@ });

src/tasks/nlp/chatCompletionStream.ts

@@ -14,5 +14,5 @@ import type { BaseArgs, Options } from "../../types";
 		...options,
-		taskHint: "text-generation",
+		task: "text-generation",
 		chatCompletion: true,
 	});
 }

src/tasks/nlp/featureExtraction.ts

@@ -0,1 +1,2 @@
+import type { FeatureExtractionInput } from "@huggingface/tasks";
 import { InferenceOutputError } from "../../lib/InferenceOutputError";
@@ -5,11 +6,3 @@ import type { BaseArgs, Options } from "../../types";
 
-export type FeatureExtractionArgs = BaseArgs & {
-	/**
-	 *  The inputs is a string or a list of strings to get the features from.
-	 *
-	 *  inputs: "That is a happy person",
-	 *
-	 */
-	inputs: string | string[];
-};
+export type FeatureExtractionArgs = BaseArgs & FeatureExtractionInput;
 
@@ -30,3 +23,3 @@ /**
 		...options,
-		taskHint: "feature-extraction",
+		task: "feature-extraction",
 	});
@@ -33,0 +26,0 @@ let isValidOutput = true;

src/tasks/nlp/fillMask.ts

@@ -14,3 +14,3 @@ import type { FillMaskInput, FillMaskOutput } from "@huggingface/tasks";
 		...options,
-		taskHint: "fill-mask",
+		task: "fill-mask",
 	});
@@ -17,0 +17,0 @@ const isValidOutput =

src/tasks/nlp/questionAnswering.ts

@@ -17,3 +17,3 @@ import type { QuestionAnsweringInput, QuestionAnsweringOutput } from "@huggingface/tasks";
 		...options,
-		taskHint: "question-answering",
+		task: "question-answering",
 	});
@@ -20,0 +20,0 @@ const isValidOutput = Array.isArray(res)

src/tasks/nlp/sentenceSimilarity.ts

@@ -18,3 +18,3 @@ import type { SentenceSimilarityInput, SentenceSimilarityOutput } from "@huggingface/tasks";
 		...options,
-		taskHint: "sentence-similarity",
+		task: "sentence-similarity",
 	});
@@ -21,0 +21,0 @@

src/tasks/nlp/summarization.ts

@@ -14,3 +14,3 @@ import type { SummarizationInput, SummarizationOutput } from "@huggingface/tasks";
 		...options,
-		taskHint: "summarization",
+		task: "summarization",
 	});
@@ -17,0 +17,0 @@ const isValidOutput = Array.isArray(res) && res.every((x) => typeof x?.summary_text === "string");

src/tasks/nlp/tableQuestionAnswering.ts

@@ -17,3 +17,3 @@ import type { TableQuestionAnsweringInput, TableQuestionAnsweringOutput } from "@huggingface/tasks";
 		...options,
-		taskHint: "table-question-answering",
+		task: "table-question-answering",
 	});
@@ -20,0 +20,0 @@ const isValidOutput = Array.isArray(res) ? res.every((elem) => validate(elem)) : validate(res);

src/tasks/nlp/textClassification.ts

@@ -18,3 +18,3 @@ import type { TextClassificationInput, TextClassificationOutput } from "@huggingface/tasks";
 			...options,
-			taskHint: "text-classification",
+			task: "text-classification",
 		})
@@ -21,0 +21,0 @@ )?.[0];

src/tasks/nlp/textGeneration.ts

@@ -42,3 +42,3 @@ import type {
 			...options,
-			taskHint: "text-generation",
+			task: "text-generation",
 		});
@@ -67,3 +67,3 @@ const isValidOutput =
 			...options,
-			taskHint: "text-generation",
+			task: "text-generation",
 		});
@@ -83,3 +83,3 @@ const isValidOutput =
 				...options,
-				taskHint: "text-generation",
+				task: "text-generation",
 			})
@@ -86,0 +86,0 @@ );

src/tasks/nlp/textGenerationStream.ts

@@ -94,4 +94,4 @@ import type { TextGenerationInput } from "@huggingface/tasks";
 		...options,
-		taskHint: "text-generation",
+		task: "text-generation",
 	});
 }

src/tasks/nlp/tokenClassification.ts

@@ -19,3 +19,3 @@ import type { TokenClassificationInput, TokenClassificationOutput } from "@huggingface/tasks";
 			...options,
-			taskHint: "token-classification",
+			task: "token-classification",
 		})
@@ -22,0 +22,0 @@ );

src/tasks/nlp/translation.ts

@@ -13,3 +13,3 @@ import type { TranslationInput, TranslationOutput } from "@huggingface/tasks";
 		...options,
-		taskHint: "translation",
+		task: "translation",
 	});
@@ -16,0 +16,0 @@ const isValidOutput = Array.isArray(res) && res.every((x) => typeof x?.translation_text === "string");

src/tasks/nlp/zeroShotClassification.ts

@@ -19,3 +19,3 @@ import type { ZeroShotClassificationInput, ZeroShotClassificationOutput } from "@huggingface/tasks";
 			...options,
-			taskHint: "zero-shot-classification",
+			task: "zero-shot-classification",
 		})
@@ -22,0 +22,0 @@ );

src/tasks/tabular/tabularClassification.ts

@@ -30,3 +30,3 @@ import { InferenceOutputError } from "../../lib/InferenceOutputError";
 		...options,
-		taskHint: "tabular-classification",
+		task: "tabular-classification",
 	});
@@ -33,0 +33,0 @@ const isValidOutput = Array.isArray(res) && res.every((x) => typeof x === "number");

src/tasks/tabular/tabularRegression.ts

@@ -30,3 +30,3 @@ import { InferenceOutputError } from "../../lib/InferenceOutputError";
 		...options,
-		taskHint: "tabular-regression",
+		task: "tabular-regression",
 	});
@@ -33,0 +33,0 @@ const isValidOutput = Array.isArray(res) && res.every((x) => typeof x === "number");

src/types.ts

@@ -1,2 +0,2 @@
-import type { ChatCompletionInput, PipelineType } from "@huggingface/tasks";
+import type { ChatCompletionInput, FeatureExtractionInput, PipelineType } from "@huggingface/tasks";
 
@@ -89,5 +89,31 @@ /**
 		| ChatCompletionInput
+		| FeatureExtractionInput
 	) & {
 		parameters?: Record<string, unknown>;
-		accessToken?: string;
 	};
+
+export interface ProviderConfig {
+	baseUrl: string;
+	makeBody: (params: BodyParams) => Record<string, unknown>;
+	makeHeaders: (params: HeaderParams) => Record<string, string>;
+	makeUrl: (params: UrlParams) => string;
+}
+
+export interface HeaderParams {
+	accessToken?: string;
+	authMethod: "none" | "hf-token" | "credentials-include" | "provider-key";
+}
+
+export interface UrlParams {
+	baseUrl: string;
+	model: string;
+	task?: InferenceTask;
+	chatCompletion?: boolean;
+}
+
+export interface BodyParams {
+	args: Record<string, unknown>;
+	chatCompletion?: boolean;
+	model: string;
+	task?: InferenceTask;
+}

No alert changes

Improved metrics

Total package byte prevSize

304419

increased by3.29%

Number of package files

211

increased by1.44%

Lines of code

6788

increased by7.13%

Dependency changes

• Updated@huggingface/tasks@^0.16.3