Spaces:
Running
Running
File size: 17,319 Bytes
ca97aa9 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 |
/**
* @file Custom data structures.
*
* These are only used internally, meaning an end-user shouldn't
* need to access anything here.
*
* @module utils/data-structures
*/
/**
* Efficient Heap-based Implementation of a Priority Queue.
* It uses an array-based binary heap, where the root is at index `0`, and the
* children of node `i` are located at indices `2i + 1` and `2i + 2`, respectively.
*
* Adapted from the following sources:
* - https://stackoverflow.com/a/42919752/13989043 (original)
* - https://github.com/belladoreai/llama-tokenizer-js (minor improvements)
*/
export class PriorityQueue {
/**
* Create a new PriorityQueue.
* @param {function(any, any): boolean} comparator Comparator function to determine priority. Defaults to a MaxHeap.
*/
constructor(comparator = (a, b) => a > b, maxSize = Infinity) {
this._heap = [];
this._comparator = comparator;
this._maxSize = maxSize;
}
/**
* The size of the queue
*/
get size() {
return this._heap.length;
}
/**
* Check if the queue is empty.
* @returns {boolean} `true` if the queue is empty, `false` otherwise.
*/
isEmpty() {
return this.size === 0;
}
/**
* Return the element with the highest priority in the queue.
* @returns {any} The highest priority element in the queue.
*/
peek() {
return this._heap[0];
}
/**
* Add one or more elements to the queue.
* @param {...any} values The values to push into the queue.
* @returns {number} The new size of the queue.
*/
push(...values) {
return this.extend(values);
}
/**
* Add multiple elements to the queue.
* @param {any[]} values The values to push into the queue.
* @returns {number} The new size of the queue.
*/
extend(values) {
for (const value of values) {
if (this.size < this._maxSize) {
this._heap.push(value);
this._siftUp();
} else {
// Get index of value with the lowest priority
const smallest = this._smallest();
// If the new value has higher priority than the smallest value in the heap
// then replace the smallest value with the new value and update the heap
if (this._comparator(value, this._heap[smallest])) {
this._heap[smallest] = value;
this._siftUpFrom(smallest);
}
}
}
return this.size;
}
/**
* Remove and return the element with the highest priority in the queue.
* @returns {any} The element with the highest priority in the queue.
*/
pop() {
const poppedValue = this.peek();
const bottom = this.size - 1;
if (bottom > 0) {
this._swap(0, bottom);
}
this._heap.pop();
this._siftDown();
return poppedValue;
}
/**
* Replace the element with the highest priority in the queue with a new value.
* @param {*} value The new value.
* @returns {*} The replaced value.
*/
replace(value) {
const replacedValue = this.peek();
this._heap[0] = value;
this._siftDown();
return replacedValue;
}
/**
* Compute the index for the parent of the node at index `i`.
* @param {number} i The index of the node to get the parent of.
* @returns {number} The index of the parent node.
* @private
*/
_parent(i) {
return ((i + 1) >>> 1) - 1;
}
/**
* Compute the index for the left child of the node at index `i`.
* @param {number} i The index of the node to get the left child of.
* @returns {number} The index of the left child.
* @private
*/
_left(i) {
return (i << 1) + 1;
}
/**
* Compute the index for the right child of the node at index `i`.
* @param {number} i The index of the node to get the right child of.
* @returns {number} The index of the right child.
* @private
*/
_right(i) {
return (i + 1) << 1;
}
/**
* Check if the element at index `i` is greater than the element at index `j`.
* @param {number} i The index of the first element to compare.
* @param {number} j The index of the second element to compare.
* @returns {boolean} `true` if the element at index `i` is greater than the element at index `j`, `false` otherwise.
* @private
*/
_greater(i, j) {
return this._comparator(this._heap[i], this._heap[j]);
}
/**
* Swap the elements at indices `i` and `j`.
* @param {number} i The index of the first element to swap.
* @param {number} j The index of the second element to swap.
* @private
*/
_swap(i, j) {
const temp = this._heap[i];
this._heap[i] = this._heap[j];
this._heap[j] = temp;
}
/**
* Maintain the heap property by updating positions in the heap,
* starting at the last element and moving up the heap.
* @private
*/
_siftUp() {
this._siftUpFrom(this.size - 1);
}
/**
* Helper function to sift up from a given node.
* @param {number} node The index of the node to start sifting up from.
*/
_siftUpFrom(node) {
while (node > 0 && this._greater(node, this._parent(node))) {
this._swap(node, this._parent(node));
node = this._parent(node);
}
}
/**
* Maintain the heap property by updating positions in the heap,
* starting at the first element and moving down the heap.
* @private
*/
_siftDown() {
let node = 0;
while (
(this._left(node) < this.size && this._greater(this._left(node), node)) ||
(this._right(node) < this.size && this._greater(this._right(node), node))
) {
const maxChild = (this._right(node) < this.size && this._greater(this._right(node), this._left(node)))
? this._right(node)
: this._left(node);
this._swap(node, maxChild);
node = maxChild;
}
}
/**
* Get the index of the smallest element in the heap. Since we use an array-based heap,
* the index can be computed without needing to traverse the heap.
* @private
*/
_smallest() {
return (2 ** (Math.floor(Math.log2(this.size))) - 1);
}
}
/**
* A trie structure to efficiently store and search for strings.
*/
export class CharTrie {
constructor() {
this.root = CharTrieNode.default();
}
/**
* Adds one or more `texts` to the trie.
* @param {string[]} texts The strings to add to the trie.
*/
extend(texts) {
for (const text of texts) {
this.push(text);
}
}
/**
* Adds text to the trie.
* @param {string} text The string to add to the trie.
*/
push(text) {
let node = this.root;
for (const ch of text) {
let child = node.children.get(ch);
if (child === undefined) {
child = CharTrieNode.default();
node.children.set(ch, child);
}
node = child;
}
node.isLeaf = true;
}
/**
* Searches the trie for all strings with a common prefix of `text`.
* @param {string} text The common prefix to search for.
* @yields {string} Each string in the trie that has `text` as a prefix.
*/
*commonPrefixSearch(text) {
let node = this.root;
if (node === undefined) return;
let prefix = "";
for (const ch of text) {
prefix += ch;
node = node.children.get(ch);
if (node === undefined) return;
if (node.isLeaf) {
yield prefix;
}
}
}
}
/**
* Represents a node in a character trie.
*/
class CharTrieNode {
/**
* Create a new CharTrieNode.
* @param {boolean} isLeaf Whether the node is a leaf node or not.
* @param {Map<string, CharTrieNode>} children A map containing the node's children, where the key is a character and the value is a `CharTrieNode`.
*/
constructor(isLeaf, children) {
this.isLeaf = isLeaf;
this.children = children;
}
/**
* Returns a new `CharTrieNode` instance with default values.
* @returns {CharTrieNode} A new `CharTrieNode` instance with `isLeaf` set to `false` and an empty `children` map.
*/
static default() {
return new CharTrieNode(false, new Map());
}
}
/**
* A lattice data structure to be used for tokenization.
*/
export class TokenLattice {
/**
* Creates a new TokenLattice instance.
*
* @param {string} sentence The input sentence to be tokenized.
* @param {number} bosTokenId The beginning-of-sequence token ID.
* @param {number} eosTokenId The end-of-sequence token ID.
*/
constructor(sentence, bosTokenId, eosTokenId) {
this.chars = Array.from(sentence);
this.len = this.chars.length;
this.bosTokenId = bosTokenId;
this.eosTokenId = eosTokenId;
this.nodes = [];
this.beginNodes = Array.from({ length: this.len + 1 }, () => []);
this.endNodes = Array.from({ length: this.len + 1 }, () => []);
const bos = new TokenLatticeNode(this.bosTokenId, 0, 0, 0, 0.0);
const eos = new TokenLatticeNode(this.eosTokenId, 1, this.len, 0, 0.0);
this.nodes.push(bos.clone());
this.nodes.push(eos.clone());
this.beginNodes[this.len].push(eos);
this.endNodes[0].push(bos);
}
/**
* Inserts a new token node into the token lattice.
*
* @param {number} pos The starting position of the token.
* @param {number} length The length of the token.
* @param {number} score The score of the token.
* @param {number} tokenId The token ID of the token.
*/
insert(pos, length, score, tokenId) {
const nodeId = this.nodes.length;
const node = new TokenLatticeNode(tokenId, nodeId, pos, length, score);
this.beginNodes[pos].push(node);
this.endNodes[pos + length].push(node);
this.nodes.push(node);
}
/**
* Implements the Viterbi algorithm to compute the most likely sequence of tokens.
*
* @returns {TokenLatticeNode[]} The most likely sequence of tokens.
*/
viterbi() {
const len = this.len;
let pos = 0;
while (pos <= len) {
if (this.beginNodes[pos].length == 0) {
return [];
}
for (let rnode of this.beginNodes[pos]) {
rnode.prev = null;
let bestScore = 0.0;
let bestNode = null;
for (let lnode of this.endNodes[pos]) {
const score = lnode.backtraceScore + rnode.score;
if (bestNode === null || score > bestScore) {
bestNode = lnode.clone();
bestScore = score;
}
}
if (bestNode !== null) {
rnode.prev = bestNode;
rnode.backtraceScore = bestScore;
} else {
return [];
}
}
++pos;
}
const results = [];
const root = this.beginNodes[len][0];
const prev = root.prev;
if (prev === null) {
return [];
}
let node = prev.clone();
while (node.prev !== null) {
results.push(node.clone());
const n = node.clone();
node = n.prev.clone();
}
results.reverse();
return results;
}
/**
* @param {TokenLatticeNode} node
* @returns {string} The array of nodes representing the most likely sequence of tokens.
*/
piece(node) {
return this.chars.slice(node.pos, node.pos + node.length).join('');
}
/**
* @returns {string[]} The most likely sequence of tokens.
*/
tokens() {
const nodes = this.viterbi();
return nodes.map(x => this.piece(x));
}
/**
* @returns {number[]} The most likely sequence of token ids.
*/
tokenIds() {
const nodes = this.viterbi();
return nodes.map(x => x.tokenId);
}
}
class TokenLatticeNode {
/**
* Represents a node in a token lattice for a given sentence.
* @param {number} tokenId The ID of the token associated with this node.
* @param {number} nodeId The ID of this node.
* @param {number} pos The starting position of the token in the sentence.
* @param {number} length The length of the token.
* @param {number} score The score associated with the token.
*/
constructor(tokenId, nodeId, pos, length, score) {
this.tokenId = tokenId;
this.nodeId = nodeId;
this.pos = pos;
this.length = length;
this.score = score;
this.prev = null;
this.backtraceScore = 0.0;
}
/**
* Returns a clone of this node.
* @returns {TokenLatticeNode} A clone of this node.
*/
clone() {
const n = new TokenLatticeNode(this.tokenId, this.nodeId, this.pos, this.length, this.score);
n.prev = this.prev;
n.backtraceScore = this.backtraceScore;
return n;
}
}
/**
* A data structure which uses a trie to split a string into tokens based on a dictionary.
* It can also use a regular expression to preprocess the input text before splitting.
*
* NOTE: To ensure multi-byte characters are handled correctly, we operate at byte-level instead of character-level.
*/
export class DictionarySplitter {
/**
* @param {string[]} dictionary The dictionary of words to use for splitting.
*/
constructor(dictionary) {
this.trie = this._buildTrie(dictionary);
}
/**
* Builds a trie from the given dictionary.
* @param {string[]} dictionary The dictionary of words to build the trie from.
* @returns {Object} The root node of the trie.
* @private
*/
_buildTrie(dictionary) {
const trie = Object.create(null);
for (const word of dictionary) {
let node = trie;
for (let i = 0; i < word.length; ++i) {
node = (node[word[i]] ??= Object.create(null));
}
node.end = word;
}
return trie;
}
/**
* Splits the input text into tokens based on the dictionary.
* @param {string} text The input text to split.
* @returns {string[]} An array of tokens.
*/
split(text) {
const result = [];
const n = text.length;
let start = 0;
let i = 0;
while (i < n) {
let node = this.trie;
let match = null;
let j = i;
while (j < n && (node = node[text[j]])) {
if (node.end) {
// Always keep the last (i.e., longest) match.
match = node.end;
}
++j;
}
if (match) {
if (i > start) {
result.push(text.slice(start, i));
}
result.push(match);
i += match.length;
start = i;
} else {
++i;
}
}
if (start < n) {
result.push(text.slice(start));
}
return result;
}
}
/**
* A simple Least Recently Used (LRU) cache implementation in JavaScript.
* This cache stores key-value pairs and evicts the least recently used item
* when the capacity is exceeded.
*/
export class LRUCache {
/**
* Creates an LRUCache instance.
* @param {number} capacity The maximum number of items the cache can hold.
*/
constructor(capacity) {
this.capacity = capacity;
this.cache = new Map();
}
/**
* Retrieves the value associated with the given key and marks the key as recently used.
* @param {any} key The key to retrieve.
* @returns {any} The value associated with the key, or undefined if the key does not exist.
*/
get(key) {
if (!this.cache.has(key)) return undefined;
const value = this.cache.get(key);
this.cache.delete(key);
this.cache.set(key, value);
return value;
}
/**
* Inserts or updates the key-value pair in the cache.
* If the key already exists, it is updated and marked as recently used.
* If the cache exceeds its capacity, the least recently used item is evicted.
* @param {any} key The key to add or update.
* @param {any} value The value to associate with the key.
*/
put(key, value) {
if (this.cache.has(key)) {
this.cache.delete(key);
}
this.cache.set(key, value);
if (this.cache.size > this.capacity) {
this.cache.delete(this.cache.keys().next().value);
}
}
/**
* Clears the cache.
*/
clear() {
this.cache.clear();
}
}
|