coyotte508 HF staff vvmnnnkv commited on
Commit
181c0bd
0 Parent(s):

Duplicate from vvmnnnkv/doc-vis-qa

Browse files

Co-authored-by: Vova Manannikov <[email protected]>

Files changed (3) hide show
  1. README.md +17 -0
  2. index.html +175 -0
  3. index.mjs +616 -0
README.md ADDED
@@ -0,0 +1,17 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ title: Document and visual question answering
3
+ emoji: ❓
4
+ colorFrom: pink
5
+ colorTo: indigo
6
+ sdk: static
7
+ pinned: false
8
+ license: mit
9
+ description: Showcase document & visual question answering using huggingface.js
10
+ duplicated_from: vvmnnnkv/doc-vis-qa
11
+ ---
12
+
13
+ Showcase document & visual question answering using the `@huggingface/inference` JS lib.
14
+
15
+ Default models for inference:
16
+ * Documents: https://huggingface.co/impira/layoutlm-document-qa
17
+ * Images: https://huggingface.co/dandelin/vilt-b32-finetuned-vqa
index.html ADDED
@@ -0,0 +1,175 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <!DOCTYPE html>
2
+ <html>
3
+ <head>
4
+ <meta charset="UTF-8"/>
5
+ <meta name="viewport" content="width=device-width, initial-scale=1.0"/>
6
+ <script src="https://cdn.tailwindcss.com"></script>
7
+ <!-- polyfill for firefox + import maps -->
8
+ <script src="https://unpkg.com/[email protected]/dist/es-module-shims.js"></script>
9
+ <script type="importmap">
10
+ {
11
+ "imports": {
12
+ "@huggingface/inference": "./index.mjs"
13
+ }
14
+ }
15
+ </script>
16
+ </head>
17
+ <body>
18
+ <form class="w-[90%] mx-auto pt-8" onsubmit="launch(); return false;">
19
+ <h1 class="text-3xl font-bold">
20
+ <span
21
+ class="bg-clip-text text-transparent bg-gradient-to-r from-pink-500 to-violet-500"
22
+ >
23
+ Document & visual question answering demo with
24
+ <a href="https://github.com/huggingface/huggingface.js">
25
+ <kbd>@huggingface/inference</kbd>
26
+ </a>
27
+ </span>
28
+ </h1>
29
+
30
+ <p class="mt-8">
31
+ First, input your token if you have one! Otherwise, you may encounter
32
+ rate limiting. You can create a token for free at
33
+ <a
34
+ target="_blank"
35
+ href="https://huggingface.co/settings/tokens"
36
+ class="underline text-blue-500"
37
+ >hf.co/settings/tokens</a
38
+ >
39
+ </p>
40
+
41
+ <input
42
+ type="text"
43
+ id="token"
44
+ class="rounded border-2 border-blue-500 shadow-md px-3 py-2 w-96 mt-6"
45
+ placeholder="token (optional)"
46
+ />
47
+
48
+ <p class="mt-8">
49
+ Pick the model type and the model you want to run. Check out models for
50
+ <a
51
+ href="https://huggingface.co/tasks/document-question-answering"
52
+ class="underline text-blue-500"
53
+ target="_blank"
54
+ >
55
+ document</a
56
+ > and
57
+ <a
58
+ href="https://huggingface.co/tasks/visual-question-answering"
59
+ class="underline text-blue-500"
60
+ target="_blank"
61
+ >image</a> question answering.
62
+ </p>
63
+
64
+ <div class="space-x-2 flex text-sm mt-8">
65
+ <label>
66
+ <input class="sr-only peer" name="type" type="radio" value="document" onclick="update_model(this.value)" checked />
67
+ <div class="px-3 py-3 rounded-lg shadow-md flex items-center justify-center text-slate-700 bg-gradient-to-r peer-checked:font-semibold peer-checked:from-pink-500 peer-checked:to-violet-500 peer-checked:text-white">
68
+ Document
69
+ </div>
70
+ </label>
71
+ <label>
72
+ <input class="sr-only peer" name="type" type="radio" value="image" onclick="update_model(this.value)" />
73
+ <div class="px-3 py-3 rounded-lg shadow-md flex items-center justify-center text-slate-700 bg-gradient-to-r peer-checked:font-semibold peer-checked:from-pink-500 peer-checked:to-violet-500 peer-checked:text-white">
74
+ Image
75
+ </div>
76
+ </label>
77
+ </div>
78
+
79
+ <input
80
+ id="model"
81
+ class="rounded border-2 border-blue-500 shadow-md px-3 py-2 w-96 mt-6"
82
+ value="impira/layoutlm-document-qa"
83
+ required
84
+ />
85
+
86
+ <p class="mt-8">The input image</p>
87
+
88
+ <input type="file" required accept="image/*"
89
+ class="rounded border-blue-500 shadow-md px-3 py-2 w-96 mt-6 block"
90
+ rows="5"
91
+ id="image"
92
+ />
93
+
94
+ <p class="mt-8">The question</p>
95
+
96
+ <input
97
+ type="text"
98
+ id="question"
99
+ class="rounded border-2 border-blue-500 shadow-md px-3 py-2 w-96 mt-6"
100
+ required
101
+ />
102
+
103
+ <button
104
+ id="submit"
105
+ class="my-8 bg-green-500 rounded py-3 px-5 text-white shadow-md disabled:bg-slate-300"
106
+ >
107
+ Run
108
+ </button>
109
+
110
+ <p class="text-gray-400 text-sm">Output logs</p>
111
+ <div id="logs" class="bg-gray-100 rounded p-3 mb-8 text-sm">
112
+ Output will be here
113
+ </div>
114
+
115
+ <p>Check out the <a class="underline text-blue-500"
116
+ href="#"
117
+ target="_blank">source code</a></p>
118
+ </form>
119
+
120
+ <script type="module">
121
+ import {HfInference} from "@huggingface/inference";
122
+
123
+ const default_models = {
124
+ "document": "impira/layoutlm-document-qa",
125
+ "image": "dandelin/vilt-b32-finetuned-vqa",
126
+ };
127
+
128
+ let running = false;
129
+
130
+ async function launch() {
131
+ if (running) {
132
+ return;
133
+ }
134
+ running = true;
135
+ try {
136
+ const hf = new HfInference(
137
+ document.getElementById("token").value.trim() || undefined
138
+ );
139
+ const model = document.getElementById("model").value.trim();
140
+ const model_type = document.querySelector("[name=type]:checked").value;
141
+ const image = document.getElementById("image").files[0];
142
+ const question = document.getElementById("question").value.trim();
143
+ document.getElementById("logs").textContent = "";
144
+
145
+ const method = model_type === "document" ? hf.documentQuestionAnswering : hf.visualQuestionAnswering;
146
+ const {answer, score} = await method({model, inputs: {
147
+ image, question
148
+ }});
149
+
150
+ document.getElementById("logs").textContent = answer + ": " + score;
151
+ } catch (err) {
152
+ alert("Error: " + err.message);
153
+ } finally {
154
+ running = false;
155
+ }
156
+ }
157
+
158
+ window.launch = launch;
159
+
160
+ window.update_model = (model_type) => {
161
+ const model_input = document.getElementById("model");
162
+ const cur_model = model_input.value.trim();
163
+ let new_model = "";
164
+ if (
165
+ model_type === "document" && cur_model === default_models["image"]
166
+ || model_type === "image" && cur_model === default_models["document"]
167
+ || cur_model === ""
168
+ ) {
169
+ new_model = default_models[model_type];
170
+ }
171
+ model_input.value = new_model;
172
+ };
173
+ </script>
174
+ </body>
175
+ </html>
index.mjs ADDED
@@ -0,0 +1,616 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ var __defProp = Object.defineProperty;
2
+ var __export = (target, all) => {
3
+ for (var name in all)
4
+ __defProp(target, name, { get: all[name], enumerable: true });
5
+ };
6
+
7
+ // src/tasks/index.ts
8
+ var tasks_exports = {};
9
+ __export(tasks_exports, {
10
+ audioClassification: () => audioClassification,
11
+ automaticSpeechRecognition: () => automaticSpeechRecognition,
12
+ conversational: () => conversational,
13
+ documentQuestionAnswering: () => documentQuestionAnswering,
14
+ featureExtraction: () => featureExtraction,
15
+ fillMask: () => fillMask,
16
+ imageClassification: () => imageClassification,
17
+ imageSegmentation: () => imageSegmentation,
18
+ imageToText: () => imageToText,
19
+ objectDetection: () => objectDetection,
20
+ questionAnswering: () => questionAnswering,
21
+ request: () => request,
22
+ sentenceSimilarity: () => sentenceSimilarity,
23
+ streamingRequest: () => streamingRequest,
24
+ summarization: () => summarization,
25
+ tableQuestionAnswering: () => tableQuestionAnswering,
26
+ textClassification: () => textClassification,
27
+ textGeneration: () => textGeneration,
28
+ textGenerationStream: () => textGenerationStream,
29
+ textToImage: () => textToImage,
30
+ tokenClassification: () => tokenClassification,
31
+ translation: () => translation,
32
+ visualQuestionAnswering: () => visualQuestionAnswering,
33
+ zeroShotClassification: () => zeroShotClassification
34
+ });
35
+
36
+ // src/lib/makeRequestOptions.ts
37
+ var HF_INFERENCE_API_BASE_URL = "https://api-inference.huggingface.co/models/";
38
+ function makeRequestOptions(args, options) {
39
+ const { model, accessToken, ...otherArgs } = args;
40
+ const headers = {};
41
+ if (accessToken) {
42
+ headers["Authorization"] = `Bearer ${accessToken}`;
43
+ }
44
+ const binary = "data" in args && !!args.data;
45
+ if (!binary) {
46
+ headers["Content-Type"] = "application/json";
47
+ } else {
48
+ if (options?.wait_for_model) {
49
+ headers["X-Wait-For-Model"] = "true";
50
+ }
51
+ if (options?.use_cache === false) {
52
+ headers["X-Use-Cache"] = "false";
53
+ }
54
+ if (options?.dont_load_model) {
55
+ headers["X-Load-Model"] = "0";
56
+ }
57
+ }
58
+ const url = /^http(s?):/.test(model) || model.startsWith("/") ? model : `${HF_INFERENCE_API_BASE_URL}${model}`;
59
+ const info = {
60
+ headers,
61
+ method: "POST",
62
+ body: binary ? args.data : JSON.stringify({
63
+ ...otherArgs,
64
+ options
65
+ }),
66
+ credentials: options?.includeCredentials ? "include" : "same-origin"
67
+ };
68
+ return { url, info };
69
+ }
70
+
71
+ // src/tasks/custom/request.ts
72
+ async function request(args, options) {
73
+ const { url, info } = makeRequestOptions(args, options);
74
+ const response = await fetch(url, info);
75
+ if (options?.retry_on_error !== false && response.status === 503 && !options?.wait_for_model) {
76
+ return request(args, {
77
+ ...options,
78
+ wait_for_model: true
79
+ });
80
+ }
81
+ if (!response.ok) {
82
+ if (response.headers.get("Content-Type")?.startsWith("application/json")) {
83
+ const output = await response.json();
84
+ if (output.error) {
85
+ throw new Error(output.error);
86
+ }
87
+ }
88
+ throw new Error("An error occurred while fetching the blob");
89
+ }
90
+ if (response.headers.get("Content-Type")?.startsWith("application/json")) {
91
+ return await response.json();
92
+ }
93
+ return await response.blob();
94
+ }
95
+
96
+ // src/vendor/fetch-event-source/parse.ts
97
+ function getLines(onLine) {
98
+ let buffer;
99
+ let position;
100
+ let fieldLength;
101
+ let discardTrailingNewline = false;
102
+ return function onChunk(arr) {
103
+ if (buffer === void 0) {
104
+ buffer = arr;
105
+ position = 0;
106
+ fieldLength = -1;
107
+ } else {
108
+ buffer = concat(buffer, arr);
109
+ }
110
+ const bufLength = buffer.length;
111
+ let lineStart = 0;
112
+ while (position < bufLength) {
113
+ if (discardTrailingNewline) {
114
+ if (buffer[position] === 10 /* NewLine */) {
115
+ lineStart = ++position;
116
+ }
117
+ discardTrailingNewline = false;
118
+ }
119
+ let lineEnd = -1;
120
+ for (; position < bufLength && lineEnd === -1; ++position) {
121
+ switch (buffer[position]) {
122
+ case 58 /* Colon */:
123
+ if (fieldLength === -1) {
124
+ fieldLength = position - lineStart;
125
+ }
126
+ break;
127
+ case 13 /* CarriageReturn */:
128
+ discardTrailingNewline = true;
129
+ case 10 /* NewLine */:
130
+ lineEnd = position;
131
+ break;
132
+ }
133
+ }
134
+ if (lineEnd === -1) {
135
+ break;
136
+ }
137
+ onLine(buffer.subarray(lineStart, lineEnd), fieldLength);
138
+ lineStart = position;
139
+ fieldLength = -1;
140
+ }
141
+ if (lineStart === bufLength) {
142
+ buffer = void 0;
143
+ } else if (lineStart !== 0) {
144
+ buffer = buffer.subarray(lineStart);
145
+ position -= lineStart;
146
+ }
147
+ };
148
+ }
149
+ function getMessages(onId, onRetry, onMessage) {
150
+ let message = newMessage();
151
+ const decoder = new TextDecoder();
152
+ return function onLine(line, fieldLength) {
153
+ if (line.length === 0) {
154
+ onMessage?.(message);
155
+ message = newMessage();
156
+ } else if (fieldLength > 0) {
157
+ const field = decoder.decode(line.subarray(0, fieldLength));
158
+ const valueOffset = fieldLength + (line[fieldLength + 1] === 32 /* Space */ ? 2 : 1);
159
+ const value = decoder.decode(line.subarray(valueOffset));
160
+ switch (field) {
161
+ case "data":
162
+ message.data = message.data ? message.data + "\n" + value : value;
163
+ break;
164
+ case "event":
165
+ message.event = value;
166
+ break;
167
+ case "id":
168
+ onId(message.id = value);
169
+ break;
170
+ case "retry":
171
+ const retry = parseInt(value, 10);
172
+ if (!isNaN(retry)) {
173
+ onRetry(message.retry = retry);
174
+ }
175
+ break;
176
+ }
177
+ }
178
+ };
179
+ }
180
+ function concat(a, b) {
181
+ const res = new Uint8Array(a.length + b.length);
182
+ res.set(a);
183
+ res.set(b, a.length);
184
+ return res;
185
+ }
186
+ function newMessage() {
187
+ return {
188
+ data: "",
189
+ event: "",
190
+ id: "",
191
+ retry: void 0
192
+ };
193
+ }
194
+
195
+ // src/tasks/custom/streamingRequest.ts
196
+ async function* streamingRequest(args, options) {
197
+ const { url, info } = makeRequestOptions({ ...args, stream: true }, options);
198
+ const response = await fetch(url, info);
199
+ if (options?.retry_on_error !== false && response.status === 503 && !options?.wait_for_model) {
200
+ return streamingRequest(args, {
201
+ ...options,
202
+ wait_for_model: true
203
+ });
204
+ }
205
+ if (!response.ok) {
206
+ if (response.headers.get("Content-Type")?.startsWith("application/json")) {
207
+ const output = await response.json();
208
+ if (output.error) {
209
+ throw new Error(output.error);
210
+ }
211
+ }
212
+ throw new Error(`Server response contains error: ${response.status}`);
213
+ }
214
+ if (response.headers.get("content-type") !== "text/event-stream") {
215
+ throw new Error(
216
+ `Server does not support event stream content type, it returned ` + response.headers.get("content-type")
217
+ );
218
+ }
219
+ if (!response.body) {
220
+ return;
221
+ }
222
+ const reader = response.body.getReader();
223
+ let events = [];
224
+ const onEvent = (event) => {
225
+ events.push(event);
226
+ };
227
+ const onChunk = getLines(
228
+ getMessages(
229
+ () => {
230
+ },
231
+ () => {
232
+ },
233
+ onEvent
234
+ )
235
+ );
236
+ try {
237
+ while (true) {
238
+ const { done, value } = await reader.read();
239
+ if (done)
240
+ return;
241
+ onChunk(value);
242
+ for (const event of events) {
243
+ if (event.data.length > 0) {
244
+ yield JSON.parse(event.data);
245
+ }
246
+ }
247
+ events = [];
248
+ }
249
+ } finally {
250
+ reader.releaseLock();
251
+ }
252
+ }
253
+
254
+ // src/lib/InferenceOutputError.ts
255
+ var InferenceOutputError = class extends TypeError {
256
+ constructor(message) {
257
+ super(
258
+ `Invalid inference output: ${message}. Use the 'request' method with the same parameters to do a custom call with no type checking.`
259
+ );
260
+ this.name = "InferenceOutputError";
261
+ }
262
+ };
263
+
264
+ // src/tasks/audio/audioClassification.ts
265
+ async function audioClassification(args, options) {
266
+ const res = await request(args, options);
267
+ const isValidOutput = Array.isArray(res) && res.every((x) => typeof x.label === "string" && typeof x.score === "number");
268
+ if (!isValidOutput) {
269
+ throw new InferenceOutputError("Expected Array<{label: string, score: number}>");
270
+ }
271
+ return res;
272
+ }
273
+
274
+ // src/tasks/audio/automaticSpeechRecognition.ts
275
+ async function automaticSpeechRecognition(args, options) {
276
+ const res = await request(args, options);
277
+ const isValidOutput = typeof res?.text === "string";
278
+ if (!isValidOutput) {
279
+ throw new InferenceOutputError("Expected {text: string}");
280
+ }
281
+ return res;
282
+ }
283
+
284
+ // src/tasks/cv/imageClassification.ts
285
+ async function imageClassification(args, options) {
286
+ const res = await request(args, options);
287
+ const isValidOutput = Array.isArray(res) && res.every((x) => typeof x.label === "string" && typeof x.score === "number");
288
+ if (!isValidOutput) {
289
+ throw new InferenceOutputError("Expected Array<{label: string, score: number}>");
290
+ }
291
+ return res;
292
+ }
293
+
294
+ // src/tasks/cv/imageSegmentation.ts
295
+ async function imageSegmentation(args, options) {
296
+ const res = await request(args, options);
297
+ const isValidOutput = Array.isArray(res) && res.every((x) => typeof x.label === "string" && typeof x.mask === "string" && typeof x.score === "number");
298
+ if (!isValidOutput) {
299
+ throw new InferenceOutputError("Expected Array<{label: string, mask: string, score: number}>");
300
+ }
301
+ return res;
302
+ }
303
+
304
+ // src/tasks/cv/imageToText.ts
305
+ async function imageToText(args, options) {
306
+ const res = (await request(args, options))?.[0];
307
+ if (typeof res?.generated_text !== "string") {
308
+ throw new InferenceOutputError("Expected {generated_text: string}");
309
+ }
310
+ return res;
311
+ }
312
+
313
+ // src/tasks/cv/objectDetection.ts
314
+ async function objectDetection(args, options) {
315
+ const res = await request(args, options);
316
+ const isValidOutput = Array.isArray(res) && res.every(
317
+ (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"
318
+ );
319
+ if (!isValidOutput) {
320
+ throw new InferenceOutputError(
321
+ "Expected Array<{label:string; score:number; box:{xmin:number; ymin:number; xmax:number; ymax:number}}>"
322
+ );
323
+ }
324
+ return res;
325
+ }
326
+
327
+ // src/tasks/cv/textToImage.ts
328
+ async function textToImage(args, options) {
329
+ const res = await request(args, options);
330
+ const isValidOutput = res && res instanceof Blob;
331
+ if (!isValidOutput) {
332
+ throw new InferenceOutputError("Expected Blob");
333
+ }
334
+ return res;
335
+ }
336
+
337
+ // src/tasks/nlp/conversational.ts
338
+ async function conversational(args, options) {
339
+ const res = await request(args, options);
340
+ const isValidOutput = Array.isArray(res.conversation.generated_responses) && res.conversation.generated_responses.every((x) => typeof x === "string") && Array.isArray(res.conversation.past_user_inputs) && res.conversation.past_user_inputs.every((x) => typeof x === "string") && typeof res.generated_text === "string" && Array.isArray(res.warnings) && res.warnings.every((x) => typeof x === "string");
341
+ if (!isValidOutput) {
342
+ throw new InferenceOutputError(
343
+ "Expected {conversation: {generated_responses: string[], past_user_inputs: string[]}, generated_text: string, warnings: string[]}"
344
+ );
345
+ }
346
+ return res;
347
+ }
348
+
349
+ // src/tasks/nlp/featureExtraction.ts
350
+ async function featureExtraction(args, options) {
351
+ const res = await request(args, options);
352
+ let isValidOutput = true;
353
+ if (Array.isArray(res)) {
354
+ for (const e of res) {
355
+ if (Array.isArray(e)) {
356
+ isValidOutput = e.every((x) => typeof x === "number");
357
+ if (!isValidOutput) {
358
+ break;
359
+ }
360
+ } else if (typeof e !== "number") {
361
+ isValidOutput = false;
362
+ break;
363
+ }
364
+ }
365
+ } else {
366
+ isValidOutput = false;
367
+ }
368
+ if (!isValidOutput) {
369
+ throw new InferenceOutputError("Expected Array<number[] | number>");
370
+ }
371
+ return res;
372
+ }
373
+
374
+ // src/tasks/nlp/fillMask.ts
375
+ async function fillMask(args, options) {
376
+ const res = await request(args, options);
377
+ const isValidOutput = Array.isArray(res) && res.every(
378
+ (x) => typeof x.score === "number" && typeof x.sequence === "string" && typeof x.token === "number" && typeof x.token_str === "string"
379
+ );
380
+ if (!isValidOutput) {
381
+ throw new InferenceOutputError(
382
+ "Expected Array<{score: number, sequence: string, token: number, token_str: string}>"
383
+ );
384
+ }
385
+ return res;
386
+ }
387
+
388
+ // src/tasks/nlp/questionAnswering.ts
389
+ async function questionAnswering(args, options) {
390
+ const res = await request(args, options);
391
+ const isValidOutput = typeof res?.answer === "string" && typeof res.end === "number" && typeof res.score === "number" && typeof res.start === "number";
392
+ if (!isValidOutput) {
393
+ throw new InferenceOutputError("Expected {answer: string, end: number, score: number, start: number}");
394
+ }
395
+ return res;
396
+ }
397
+
398
+ // src/tasks/nlp/sentenceSimilarity.ts
399
+ async function sentenceSimilarity(args, options) {
400
+ const res = await request(args, options);
401
+ const isValidOutput = Array.isArray(res) && res.every((x) => typeof x === "number");
402
+ if (!isValidOutput) {
403
+ throw new InferenceOutputError("Expected number[]");
404
+ }
405
+ return res;
406
+ }
407
+
408
+ // src/tasks/nlp/summarization.ts
409
+ async function summarization(args, options) {
410
+ const res = await request(args, options);
411
+ const isValidOutput = Array.isArray(res) && res.every((x) => typeof x?.summary_text === "string");
412
+ if (!isValidOutput) {
413
+ throw new InferenceOutputError("Expected Array<{summary_text: string}>");
414
+ }
415
+ return res?.[0];
416
+ }
417
+
418
+ // src/tasks/nlp/tableQuestionAnswering.ts
419
+ async function tableQuestionAnswering(args, options) {
420
+ const res = await request(args, options);
421
+ const isValidOutput = typeof res?.aggregator === "string" && typeof res.answer === "string" && Array.isArray(res.cells) && res.cells.every((x) => typeof x === "string") && Array.isArray(res.coordinates) && res.coordinates.every((coord) => Array.isArray(coord) && coord.every((x) => typeof x === "number"));
422
+ if (!isValidOutput) {
423
+ throw new InferenceOutputError(
424
+ "Expected {aggregator: string, answer: string, cells: string[], coordinates: number[][]}"
425
+ );
426
+ }
427
+ return res;
428
+ }
429
+
430
+ // src/tasks/nlp/textClassification.ts
431
+ async function textClassification(args, options) {
432
+ const res = (await request(args, options))?.[0];
433
+ const isValidOutput = Array.isArray(res) && res.every((x) => typeof x?.label === "string" && typeof x.score === "number");
434
+ if (!isValidOutput) {
435
+ throw new InferenceOutputError("Expected Array<{label: string, score: number}>");
436
+ }
437
+ return res;
438
+ }
439
+
440
+ // src/tasks/nlp/textGeneration.ts
441
+ async function textGeneration(args, options) {
442
+ const res = await request(args, options);
443
+ const isValidOutput = Array.isArray(res) && res.every((x) => typeof x?.generated_text === "string");
444
+ if (!isValidOutput) {
445
+ throw new InferenceOutputError("Expected Array<{generated_text: string}>");
446
+ }
447
+ return res?.[0];
448
+ }
449
+
450
+ // src/tasks/nlp/textGenerationStream.ts
451
+ async function* textGenerationStream(args, options) {
452
+ yield* streamingRequest(args, options);
453
+ }
454
+
455
+ // src/utils/toArray.ts
456
+ function toArray(obj) {
457
+ if (Array.isArray(obj)) {
458
+ return obj;
459
+ }
460
+ return [obj];
461
+ }
462
+
463
+ // src/tasks/nlp/tokenClassification.ts
464
+ async function tokenClassification(args, options) {
465
+ const res = toArray(await request(args, options));
466
+ const isValidOutput = Array.isArray(res) && res.every(
467
+ (x) => typeof x.end === "number" && typeof x.entity_group === "string" && typeof x.score === "number" && typeof x.start === "number" && typeof x.word === "string"
468
+ );
469
+ if (!isValidOutput) {
470
+ throw new InferenceOutputError(
471
+ "Expected Array<{end: number, entity_group: string, score: number, start: number, word: string}>"
472
+ );
473
+ }
474
+ return res;
475
+ }
476
+
477
+ // src/tasks/nlp/translation.ts
478
+ async function translation(args, options) {
479
+ const res = await request(args, options);
480
+ const isValidOutput = Array.isArray(res) && res.every((x) => typeof x?.translation_text === "string");
481
+ if (!isValidOutput) {
482
+ throw new InferenceOutputError("Expected type Array<{translation_text: string}>");
483
+ }
484
+ return res?.[0];
485
+ }
486
+
487
+ // src/tasks/nlp/zeroShotClassification.ts
488
+ async function zeroShotClassification(args, options) {
489
+ const res = toArray(
490
+ await request(args, options)
491
+ );
492
+ const isValidOutput = Array.isArray(res) && res.every(
493
+ (x) => Array.isArray(x.labels) && x.labels.every((_label) => typeof _label === "string") && Array.isArray(x.scores) && x.scores.every((_score) => typeof _score === "number") && typeof x.sequence === "string"
494
+ );
495
+ if (!isValidOutput) {
496
+ throw new InferenceOutputError("Expected Array<{labels: string[], scores: number[], sequence: string}>");
497
+ }
498
+ return res;
499
+ }
500
+
501
+ // ../shared/src/base64FromBytes.ts
502
+ function base64FromBytes(arr) {
503
+ if (globalThis.Buffer) {
504
+ return globalThis.Buffer.from(arr).toString("base64");
505
+ } else {
506
+ const bin = [];
507
+ arr.forEach((byte) => {
508
+ bin.push(String.fromCharCode(byte));
509
+ });
510
+ return globalThis.btoa(bin.join(""));
511
+ }
512
+ }
513
+
514
+ // src/tasks/multimodal/documentQuestionAnswering.ts
515
+ async function documentQuestionAnswering(args, options) {
516
+ const reqArgs = {
517
+ ...args,
518
+ inputs: {
519
+ question: args.inputs.question,
520
+ // convert Blob to base64
521
+ image: base64FromBytes(new Uint8Array(await args.inputs.image.arrayBuffer()))
522
+ }
523
+ };
524
+ const res = (await request(reqArgs, options))?.[0];
525
+ const isValidOutput = typeof res?.answer === "string" && typeof res.end === "number" && typeof res.score === "number" && typeof res.start === "number";
526
+ if (!isValidOutput) {
527
+ throw new InferenceOutputError("Expected Array<{answer: string, end: number, score: number, start: number}>");
528
+ }
529
+ return res;
530
+ }
531
+
532
+ // src/tasks/multimodal/visualQuestionAnswering.ts
533
+ async function visualQuestionAnswering(args, options) {
534
+ const reqArgs = {
535
+ ...args,
536
+ inputs: {
537
+ question: args.inputs.question,
538
+ // convert Blob to base64
539
+ image: base64FromBytes(new Uint8Array(await args.inputs.image.arrayBuffer()))
540
+ }
541
+ };
542
+ const res = (await request(reqArgs, options))?.[0];
543
+ const isValidOutput = typeof res?.answer === "string" && typeof res.score === "number";
544
+ if (!isValidOutput) {
545
+ throw new InferenceOutputError("Expected Array<{answer: string, score: number}>");
546
+ }
547
+ return res;
548
+ }
549
+
550
+ // src/HfInference.ts
551
+ var HfInference = class {
552
+ accessToken;
553
+ defaultOptions;
554
+ constructor(accessToken = "", defaultOptions = {}) {
555
+ this.accessToken = accessToken;
556
+ this.defaultOptions = defaultOptions;
557
+ for (const [name, fn] of Object.entries(tasks_exports)) {
558
+ Object.defineProperty(this, name, {
559
+ enumerable: false,
560
+ value: (params, options) => (
561
+ // eslint-disable-next-line @typescript-eslint/no-explicit-any
562
+ fn({ ...params, accessToken }, { ...defaultOptions, ...options })
563
+ )
564
+ });
565
+ }
566
+ }
567
+ /**
568
+ * Returns copy of HfInference tied to a specified endpoint.
569
+ */
570
+ endpoint(endpointUrl) {
571
+ return new HfInferenceEndpoint(endpointUrl, this.accessToken, this.defaultOptions);
572
+ }
573
+ };
574
+ var HfInferenceEndpoint = class {
575
+ constructor(endpointUrl, accessToken = "", defaultOptions = {}) {
576
+ accessToken;
577
+ defaultOptions;
578
+ for (const [name, fn] of Object.entries(tasks_exports)) {
579
+ Object.defineProperty(this, name, {
580
+ enumerable: false,
581
+ value: (params, options) => (
582
+ // eslint-disable-next-line @typescript-eslint/no-explicit-any
583
+ fn({ ...params, accessToken, model: endpointUrl }, { ...defaultOptions, ...options })
584
+ )
585
+ });
586
+ }
587
+ }
588
+ };
589
+ export {
590
+ HfInference,
591
+ HfInferenceEndpoint,
592
+ audioClassification,
593
+ automaticSpeechRecognition,
594
+ conversational,
595
+ documentQuestionAnswering,
596
+ featureExtraction,
597
+ fillMask,
598
+ imageClassification,
599
+ imageSegmentation,
600
+ imageToText,
601
+ objectDetection,
602
+ questionAnswering,
603
+ request,
604
+ sentenceSimilarity,
605
+ streamingRequest,
606
+ summarization,
607
+ tableQuestionAnswering,
608
+ textClassification,
609
+ textGeneration,
610
+ textGenerationStream,
611
+ textToImage,
612
+ tokenClassification,
613
+ translation,
614
+ visualQuestionAnswering,
615
+ zeroShotClassification
616
+ };