@huggingface/inference 3.12.1 → 3.13.1

package/dist/index.cjs

@@ -235,6 +235,43 @@ var BaseTextGenerationTask = class extends TaskProviderHelper {
   }
 };
 
+// src/utils/base64FromBytes.ts
+function base64FromBytes(arr) {
+  if (globalThis.Buffer) {
+    return globalThis.Buffer.from(arr).toString("base64");
+  } else {
+    const bin = [];
+    arr.forEach((byte) => {
+      bin.push(String.fromCharCode(byte));
+    });
+    return globalThis.btoa(bin.join(""));
+  }
+}
+
+// src/utils/pick.ts
+function pick(o, props) {
+  return Object.assign(
+    {},
+    ...props.map((prop) => {
+      if (o[prop] !== void 0) {
+        return { [prop]: o[prop] };
+      }
+    })
+  );
+}
+
+// src/utils/typedInclude.ts
+function typedInclude(arr, v) {
+  return arr.includes(v);
+}
+
+// src/utils/omit.ts
+function omit(o, props) {
+  const propsArr = Array.isArray(props) ? props : [props];
+  const letsKeep = Object.keys(o).filter((prop) => !typedInclude(propsArr, prop));
+  return pick(o, letsKeep);
+}
+
 // src/providers/hf-inference.ts
 var EQUIVALENT_SENTENCE_TRANSFORMERS_TASKS = ["feature-extraction", "sentence-similarity"];
 var HFInferenceTask = class extends TaskProviderHelper {
@@ -342,6 +379,12 @@ var HFInferenceAutomaticSpeechRecognitionTask = class extends HFInferenceTask {
   async getResponse(response) {
     return response;
   }
+  async preparePayloadAsync(args) {
+    return "data" in args ? args : {
+      ...omit(args, "inputs"),
+      data: args.inputs
+    };
+  }
 };
 var HFInferenceAudioToAudioTask = class extends HFInferenceTask {
   async getResponse(response) {
@@ -393,7 +436,9 @@ var HFInferenceImageClassificationTask = class extends HFInferenceTask {
 };
 var HFInferenceImageSegmentationTask = class extends HFInferenceTask {
   async getResponse(response) {
-    if (Array.isArray(response) && response.every((x) => typeof x.label === "string" && typeof x.mask === "string" && typeof x.score === "number")) {
+    if (Array.isArray(response) && response.every(
+      (x) => typeof x.label === "string" && typeof x.mask === "string" && (x.score === void 0 || typeof x.score === "number")
+    )) {
       return response;
     }
     throw new InferenceOutputError("Expected Array<{label: string, mask: string, score: number}>");
@@ -408,6 +453,22 @@ var HFInferenceImageToTextTask = class extends HFInferenceTask {
   }
 };
 var HFInferenceImageToImageTask = class extends HFInferenceTask {
+  async preparePayloadAsync(args) {
+    if (!args.parameters) {
+      return {
+        ...args,
+        model: args.model,
+        data: args.inputs
+      };
+    } else {
+      return {
+        ...args,
+        inputs: base64FromBytes(
+          new Uint8Array(args.inputs instanceof ArrayBuffer ? args.inputs : await args.inputs.arrayBuffer())
+        )
+      };
+    }
+  }
   async getResponse(response) {
     if (response instanceof Blob) {
       return response;
@@ -564,11 +625,6 @@ var HFInferenceTextToAudioTask = class extends HFInferenceTask {
   }
 };
 
-// src/utils/typedInclude.ts
-function typedInclude(arr, v) {
-  return arr.includes(v);
-}
-
 // src/lib/getInferenceProviderMapping.ts
 var inferenceProviderMappingCache = /* @__PURE__ */ new Map();
 async function fetchInferenceProviderMappingForModel(modelId, accessToken, options) {
@@ -654,25 +710,6 @@ function delay(ms) {
   });
 }
 
-// src/utils/pick.ts
-function pick(o, props) {
-  return Object.assign(
-    {},
-    ...props.map((prop) => {
-      if (o[prop] !== void 0) {
-        return { [prop]: o[prop] };
-      }
-    })
-  );
-}
-
-// src/utils/omit.ts
-function omit(o, props) {
-  const propsArr = Array.isArray(props) ? props : [props];
-  const letsKeep = Object.keys(o).filter((prop) => !typedInclude(propsArr, prop));
-  return pick(o, letsKeep);
-}
-
 // src/providers/black-forest-labs.ts
 var BLACK_FOREST_LABS_AI_API_BASE_URL = "https://api.us1.bfl.ai";
 var BlackForestLabsTextToImageTask = class extends TaskProviderHelper {
@@ -879,6 +916,27 @@ var FalAIAutomaticSpeechRecognitionTask = class extends FalAITask {
     }
     return { text: res.text };
   }
+  async preparePayloadAsync(args) {
+    const blob = "data" in args && args.data instanceof Blob ? args.data : "inputs" in args ? args.inputs : void 0;
+    const contentType = blob?.type;
+    if (!contentType) {
+      throw new Error(
+        `Unable to determine the input's content-type. Make sure your are passing a Blob when using provider fal-ai.`
+      );
+    }
+    if (!FAL_AI_SUPPORTED_BLOB_TYPES.includes(contentType)) {
+      throw new Error(
+        `Provider fal-ai does not support blob type ${contentType} - supported content types are: ${FAL_AI_SUPPORTED_BLOB_TYPES.join(
+          ", "
+        )}`
+      );
+    }
+    const base64audio = base64FromBytes(new Uint8Array(await blob.arrayBuffer()));
+    return {
+      ..."data" in args ? omit(args, "data") : omit(args, "inputs"),
+      audio_url: `data:${contentType};base64,${base64audio}`
+    };
+  }
 };
 var FalAITextToSpeechTask = class extends FalAITask {
   preparePayload(params) {
@@ -1054,7 +1112,7 @@ var NebiusTextToImageTask = class extends TaskProviderHelper {
       model: params.model
     };
   }
-  makeRoute(params) {
+  makeRoute() {
     return "v1/images/generations";
   }
   async getResponse(response, url, headers, outputType) {
@@ -1068,6 +1126,23 @@ var NebiusTextToImageTask = class extends TaskProviderHelper {
     throw new InferenceOutputError("Expected Nebius text-to-image response format");
   }
 };
+var NebiusFeatureExtractionTask = class extends TaskProviderHelper {
+  constructor() {
+    super("nebius", NEBIUS_API_BASE_URL);
+  }
+  preparePayload(params) {
+    return {
+      input: params.args.inputs,
+      model: params.model
+    };
+  }
+  makeRoute() {
+    return "v1/embeddings";
+  }
+  async getResponse(response) {
+    return response.data.map((item) => item.embedding);
+  }
+};
 
 // src/providers/novita.ts
 var NOVITA_API_BASE_URL = "https://api.novita.ai";
@@ -1414,7 +1489,8 @@ var PROVIDERS = {
   nebius: {
     "text-to-image": new NebiusTextToImageTask(),
     conversational: new NebiusConversationalTask(),
-    "text-generation": new NebiusTextGenerationTask()
+    "text-generation": new NebiusTextGenerationTask(),
+    "feature-extraction": new NebiusFeatureExtractionTask()
   },
   novita: {
     conversational: new NovitaConversationalTask(),
@@ -1469,7 +1545,7 @@ function getProviderHelper(provider, task) {
 
 // package.json
 var name = "@huggingface/inference";
-var version = "3.12.1";
+var version = "3.13.1";
 
 // src/lib/makeRequestOptions.ts
 var tasks = null;
@@ -1870,24 +1946,11 @@ async function audioToAudio(args, options) {
   return providerHelper.getResponse(res);
 }
 
-// src/utils/base64FromBytes.ts
-function base64FromBytes(arr) {
-  if (globalThis.Buffer) {
-    return globalThis.Buffer.from(arr).toString("base64");
-  } else {
-    const bin = [];
-    arr.forEach((byte) => {
-      bin.push(String.fromCharCode(byte));
-    });
-    return globalThis.btoa(bin.join(""));
-  }
-}
-
 // src/tasks/audio/automaticSpeechRecognition.ts
 async function automaticSpeechRecognition(args, options) {
   const provider = await resolveProvider(args.provider, args.model, args.endpointUrl);
   const providerHelper = getProviderHelper(provider, "automatic-speech-recognition");
-  const payload = await buildPayload(args);
+  const payload = await providerHelper.preparePayloadAsync(args);
   const { data: res } = await innerRequest(payload, providerHelper, {
     ...options,
     task: "automatic-speech-recognition"
@@ -1898,31 +1961,6 @@ async function automaticSpeechRecognition(args, options) {
   }
   return providerHelper.getResponse(res);
 }
-async function buildPayload(args) {
-  if (args.provider === "fal-ai") {
-    const blob = "data" in args && args.data instanceof Blob ? args.data : "inputs" in args ? args.inputs : void 0;
-    const contentType = blob?.type;
-    if (!contentType) {
-      throw new Error(
-        `Unable to determine the input's content-type. Make sure your are passing a Blob when using provider fal-ai.`
-      );
-    }
-    if (!FAL_AI_SUPPORTED_BLOB_TYPES.includes(contentType)) {
-      throw new Error(
-        `Provider fal-ai does not support blob type ${contentType} - supported content types are: ${FAL_AI_SUPPORTED_BLOB_TYPES.join(
-          ", "
-        )}`
-      );
-    }
-    const base64audio = base64FromBytes(new Uint8Array(await blob.arrayBuffer()));
-    return {
-      ..."data" in args ? omit(args, "data") : omit(args, "inputs"),
-      audio_url: `data:${contentType};base64,${base64audio}`
-    };
-  } else {
-    return preparePayload(args);
-  }
-}
 
 // src/tasks/audio/textToSpeech.ts
 async function textToSpeech(args, options) {
@@ -1968,22 +2006,8 @@ async function imageSegmentation(args, options) {
 async function imageToImage(args, options) {
   const provider = await resolveProvider(args.provider, args.model, args.endpointUrl);
   const providerHelper = getProviderHelper(provider, "image-to-image");
-  let reqArgs;
-  if (!args.parameters) {
-    reqArgs = {
-      accessToken: args.accessToken,
-      model: args.model,
-      data: args.inputs
-    };
-  } else {
-    reqArgs = {
-      ...args,
-      inputs: base64FromBytes(
-        new Uint8Array(args.inputs instanceof ArrayBuffer ? args.inputs : await args.inputs.arrayBuffer())
-      )
-    };
-  }
-  const { data: res } = await innerRequest(reqArgs, providerHelper, {
+  const payload = await providerHelper.preparePayloadAsync(args);
+  const { data: res } = await innerRequest(payload, providerHelper, {
     ...options,
     task: "image-to-image"
   });
@@ -2448,7 +2472,7 @@ const video = await client.textToVideo({
       "textToImage": '{% if provider == "fal-ai" %}\nimport fal_client\n\n{% if providerInputs.asObj.loras is defined and providerInputs.asObj.loras != none %}\nresult = fal_client.subscribe(\n    "{{ providerModelId }}",\n    arguments={\n        "prompt": {{ inputs.asObj.inputs }},\n        "loras":{{ providerInputs.asObj.loras | tojson }},\n    },\n)\n{% else %}\nresult = fal_client.subscribe(\n    "{{ providerModelId }}",\n    arguments={\n        "prompt": {{ inputs.asObj.inputs }},\n    },\n)\n{% endif %} \nprint(result)\n{% endif %} '
     },
     "huggingface_hub": {
-      "basic": 'result = client.{{ methodName }}(\n    inputs={{ inputs.asObj.inputs }},\n    model="{{ model.id }}",\n)',
+      "basic": 'result = client.{{ methodName }}(\n    {{ inputs.asObj.inputs }},\n    model="{{ model.id }}",\n)',
       "basicAudio": 'output = client.{{ methodName }}({{ inputs.asObj.inputs }}, model="{{ model.id }}")',
       "basicImage": 'output = client.{{ methodName }}({{ inputs.asObj.inputs }}, model="{{ model.id }}")',
       "conversational": 'completion = client.chat.completions.create(\n    model="{{ model.id }}",\n{{ inputs.asPythonString }}\n)\n\nprint(completion.choices[0].message) ',
@@ -2456,6 +2480,8 @@ const video = await client.textToVideo({
       "documentQuestionAnswering": 'output = client.document_question_answering(\n    "{{ inputs.asObj.image }}",\n    question="{{ inputs.asObj.question }}",\n    model="{{ model.id }}",\n) ',
       "imageToImage": '# output is a PIL.Image object\nimage = client.image_to_image(\n    "{{ inputs.asObj.inputs }}",\n    prompt="{{ inputs.asObj.parameters.prompt }}",\n    model="{{ model.id }}",\n) ',
       "importInferenceClient": 'from huggingface_hub import InferenceClient\n\nclient = InferenceClient(\n    provider="{{ provider }}",\n    api_key="{{ accessToken }}",\n{% if billTo %}\n    bill_to="{{ billTo }}",\n{% endif %}\n)',
+      "questionAnswering": 'answer = client.question_answering(\n    question="{{ inputs.asObj.question }}",\n    context="{{ inputs.asObj.context }}",\n    model="{{ model.id }}",\n) ',
+      "tableQuestionAnswering": 'answer = client.question_answering(\n    query="{{ inputs.asObj.query }}",\n    table={{ inputs.asObj.table }},\n    model="{{ model.id }}",\n) ',
       "textToImage": '# output is a PIL.Image object\nimage = client.text_to_image(\n    {{ inputs.asObj.inputs }},\n    model="{{ model.id }}",\n) ',
       "textToSpeech": '# audio is returned as bytes\naudio = client.text_to_speech(\n    {{ inputs.asObj.inputs }},\n    model="{{ model.id }}",\n) \n',
       "textToVideo": 'video = client.text_to_video(\n    {{ inputs.asObj.inputs }},\n    model="{{ model.id }}",\n) '
@@ -2527,6 +2553,10 @@ var CLIENTS = {
   python: [...PYTHON_CLIENTS],
   sh: [...SH_CLIENTS]
 };
+var CLIENTS_AUTO_POLICY = {
+  js: ["huggingface.js"],
+  python: ["huggingface_hub"]
+};
 var hasTemplate = (language, client, templateName) => templates[language]?.[client]?.[templateName] !== void 0;
 var loadTemplate = (language, client, templateName) => {
   const template = templates[language]?.[client]?.[templateName];
@@ -2593,7 +2623,7 @@ var snippetGenerator = (templateName, inputPreparationFn) => {
     }
     let providerHelper;
     try {
-      providerHelper = getProviderHelper(provider, task);
+      providerHelper = getProviderHelper(provider === "auto" ? "hf-inference" : provider, task);
     } catch (e) {
       console.error(`Failed to get provider helper for ${provider} (${task})`, e);
       return [];
@@ -2646,8 +2676,10 @@ var snippetGenerator = (templateName, inputPreparationFn) => {
       providerModelId: providerModelId ?? model.id,
       billTo: opts?.billTo
     };
+    const clients = provider === "auto" ? CLIENTS_AUTO_POLICY : CLIENTS;
     return import_tasks.inferenceSnippetLanguages.map((language) => {
-      return CLIENTS[language].map((client) => {
+      const langClients = clients[language] ?? [];
+      return langClients.map((client) => {
         if (!hasTemplate(language, client, templateName)) {
           return;
         }
@@ -2703,6 +2735,14 @@ var prepareConversationalInput = (model, opts) => {
     ...opts?.top_p ? { top_p: opts?.top_p } : void 0
   };
 };
+var prepareQuestionAnsweringInput = (model) => {
+  const data = JSON.parse((0, import_tasks.getModelInputSnippet)(model));
+  return { question: data.question, context: data.context };
+};
+var prepareTableQuestionAnsweringInput = (model) => {
+  const data = JSON.parse((0, import_tasks.getModelInputSnippet)(model));
+  return { query: data.query, table: JSON.stringify(data.table) };
+};
 var snippets = {
   "audio-classification": snippetGenerator("basicAudio"),
   "audio-to-audio": snippetGenerator("basicAudio"),
@@ -2716,12 +2756,12 @@ var snippets = {
   "image-to-image": snippetGenerator("imageToImage", prepareImageToImageInput),
   "image-to-text": snippetGenerator("basicImage"),
   "object-detection": snippetGenerator("basicImage"),
-  "question-answering": snippetGenerator("basic"),
+  "question-answering": snippetGenerator("questionAnswering", prepareQuestionAnsweringInput),
   "sentence-similarity": snippetGenerator("basic"),
   summarization: snippetGenerator("basic"),
   "tabular-classification": snippetGenerator("tabular"),
   "tabular-regression": snippetGenerator("tabular"),
-  "table-question-answering": snippetGenerator("basic"),
+  "table-question-answering": snippetGenerator("tableQuestionAnswering", prepareTableQuestionAnsweringInput),
   "text-classification": snippetGenerator("basic"),
   "text-generation": snippetGenerator("basic"),
   "text-to-audio": snippetGenerator("textToAudio"),