@seanhogg/builderforce-memory-engine 2026.6.28 → 2026.6.30
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/lm/evermind_lm.d.ts +7 -7
- package/dist/lm/evermind_lm.d.ts.map +1 -1
- package/dist/tokenizer/bpe.d.ts +11 -0
- package/dist/tokenizer/bpe.d.ts.map +1 -1
- package/dist/tokenizer/bpe.js +78 -0
- package/dist/tokenizer/bpe.js.map +1 -1
- package/package.json +5 -3
- package/src/lm/evermind_lm.ts +7 -7
- package/src/tokenizer/bpe.ts +77 -0
package/dist/lm/evermind_lm.d.ts
CHANGED
|
@@ -22,20 +22,20 @@
|
|
|
22
22
|
*/
|
|
23
23
|
import { type AdamWOptions } from "../optim/adamw.js";
|
|
24
24
|
export interface EvermindLMConfig {
|
|
25
|
-
/** Vocabulary size. */
|
|
25
|
+
/** Vocabulary size (the only required field; everything else has a default). */
|
|
26
26
|
vocabSize: number;
|
|
27
27
|
/** Model (channel) dimension. Default 64. */
|
|
28
|
-
dModel
|
|
28
|
+
dModel?: number;
|
|
29
29
|
/** Number of (conv + MoE) blocks. Default 2. */
|
|
30
|
-
numLayers
|
|
30
|
+
numLayers?: number;
|
|
31
31
|
/** Causal conv kernel width. Default 3. */
|
|
32
|
-
convKernel
|
|
32
|
+
convKernel?: number;
|
|
33
33
|
/** Hidden width of each MoE expert FFN. Default 2·dModel. */
|
|
34
|
-
hiddenDim
|
|
34
|
+
hiddenDim?: number;
|
|
35
35
|
/** Routed experts per MoE layer. Default 4. */
|
|
36
|
-
numExperts
|
|
36
|
+
numExperts?: number;
|
|
37
37
|
/** Experts activated per token. Default 2. */
|
|
38
|
-
topK
|
|
38
|
+
topK?: number;
|
|
39
39
|
/** Deterministic init seed. */
|
|
40
40
|
seed?: number;
|
|
41
41
|
}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"evermind_lm.d.ts","sourceRoot":"","sources":["../../src/lm/evermind_lm.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;GAqBG;AAIH,OAAO,EAAS,KAAK,YAAY,EAAE,MAAM,mBAAmB,CAAC;AAI7D,MAAM,WAAW,gBAAgB;IAC/B,
|
|
1
|
+
{"version":3,"file":"evermind_lm.d.ts","sourceRoot":"","sources":["../../src/lm/evermind_lm.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;GAqBG;AAIH,OAAO,EAAS,KAAK,YAAY,EAAE,MAAM,mBAAmB,CAAC;AAI7D,MAAM,WAAW,gBAAgB;IAC/B,gFAAgF;IAChF,SAAS,EAAE,MAAM,CAAC;IAClB,6CAA6C;IAC7C,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,gDAAgD;IAChD,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,2CAA2C;IAC3C,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,6DAA6D;IAC7D,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,+CAA+C;IAC/C,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,8CAA8C;IAC9C,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,+BAA+B;IAC/B,IAAI,CAAC,EAAE,MAAM,CAAC;CACf;AAED,eAAO,MAAM,iBAAiB,EAAE,QAAQ,CAAC,IAAI,CAAC,gBAAgB,EAAE,MAAM,GAAG,WAAW,CAAC,CAOpF,CAAC;AAEF,eAAO,MAAM,eAAe,aAAa,CAAC;AAG1C,UAAU,YAAY;IACpB,CAAC,EAAE,YAAY,CAAC;IAChB,KAAK,EAAE;QAAE,OAAO,EAAE,MAAM,EAAE,CAAC;QAAC,KAAK,EAAE,MAAM,EAAE,CAAC;QAAC,KAAK,EAAE,YAAY,CAAA;KAAE,CAAC;IACnE,SAAS,EAAE,YAAY,CAAC;IACxB,OAAO,EAAE,YAAY,CAAC;IACtB,SAAS,EAAE,YAAY,EAAE,CAAC;IAC1B,SAAS,EAAE,YAAY,EAAE,CAAC;IAC1B,OAAO,EAAE,YAAY,EAAE,CAAC;CACzB;AAED,UAAU,UAAU;IAClB,OAAO,EAAE,YAAY,EAAE,CAAC;IACxB,UAAU,EAAE,YAAY,EAAE,CAAC;IAC3B,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,SAAS,EAAE,YAAY,EAAE,CAAC;IAC1B,MAAM,EAAE,MAAM,EAAE,CAAC;IACjB,QAAQ,EAAE,YAAY,EAAE,CAAC;CAC1B;AAED,UAAU,YAAY;IACpB,MAAM,EAAE,MAAM,EAAE,CAAC;IACjB,MAAM,EAAE,UAAU,EAAE,CAAC;IACrB,MAAM,EAAE,YAAY,EAAE,CAAC;CACxB;AAED,yFAAyF;AACzF,MAAM,WAAW,SAAS;IACxB,MAAM,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,EAAE,CAAC;IAC/B,MAAM,CAAC,GAAG,EAAE,MAAM,EAAE,GAAG,MAAM,CAAC;CAC/B;AAED,MAAM,WAAW,iBAAiB;IAChC,YAAY,EAAE,MAAM,CAAC;IACrB,oEAAoE;IACpE,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,mEAAmE;IACnE,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,sDAAsD;IACtD,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAED,qBAAa,UAAU;IACrB,QAAQ,CAAC,MAAM,EAAE,QAAQ,CAAC,IAAI,CAAC,gBAAgB,EAAE,MAAM,CAAC,CAAC,CAAC;IAE1D,0EAA0E;IAC1E,GAAG,EAAE,YAAY,CAAC;IAClB,OAAO,CAAC,IAAI,CAAe;IAC3B,oEAAoE;IACpE,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAiB;IACtC,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAiB;IACvC,gEAAgE;IAChE,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAiB;IACvC,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAiB;IACxC,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAiB;IACtC,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAiB;IACvC,+BAA+B;IAC/B,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAoB;gBAE5B,MAAM,EAAE,gBAAgB;IA+DpC,kFAAkF;IAClF,OAAO,CAAC,MAAM,EAAE,MAAM,EAAE,GAAG;QAAE,MAAM,EAAE,YAAY,EAAE,CAAC;QAAC,KAAK,EAAE,YAAY,CAAA;KAAE;IA4E1E;;;;OAIG;IACH,eAAe,CAAC,MAAM,EAAE,MAAM,EAAE,GAAG,MAAM;IAqFzC;;;;;OAKG;IACH,YAAY,CAAC,MAAM,EAAE,MAAM,EAAE,KAAK,EAAE,SAAS,EAAE,IAAI,EAAE,iBAAiB,GAAG,MAAM;IAI/E,qFAAqF;IACrF,QAAQ,CAAC,MAAM,EAAE,MAAM,EAAE,EAAE,IAAI,EAAE,iBAAiB,GAAG,MAAM,EAAE;IAkB7D,8EAA8E;IAC9E,UAAU,IAAI;QAAE,IAAI,EAAE,YAAY,CAAA;KAAE,EAAE;IAStC,+DAA+D;IAC/D,SAAS,IAAI;QAAE,IAAI,EAAE,YAAY,CAAA;KAAE,EAAE;IASrC,QAAQ,IAAI,IAAI;IAUhB,uFAAuF;IACvF,aAAa,CAAC,IAAI,GAAE;QAAE,IAAI,CAAC,EAAE,OAAO,CAAA;KAAO,GAAG,WAAW;IA+BzD,kEAAkE;IAClE,WAAW,CAAC,MAAM,EAAE,WAAW,GAAG,IAAI;CA4BvC;AAED,qEAAqE;AACrE,qBAAa,iBAAiB;IAG1B,OAAO,CAAC,QAAQ,CAAC,KAAK;IACtB,OAAO,CAAC,QAAQ,CAAC,IAAI;IAHvB,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAQ;gBAEV,KAAK,EAAE,UAAU,EACjB,IAAI,GAAE,YAAY,GAAG;QAAE,MAAM,CAAC,EAAE,MAAM,CAAA;KAAO;IAIhE,sEAAsE;IACtE,GAAG,CAAC,SAAS,EAAE,MAAM,EAAE,EAAE,GAAG,MAAM,EAAE;CAiBrC"}
|
package/dist/tokenizer/bpe.d.ts
CHANGED
|
@@ -19,6 +19,17 @@ export declare class BPETokenizer {
|
|
|
19
19
|
padId: number | null;
|
|
20
20
|
constructor();
|
|
21
21
|
load(vocab: string | Record<string, number>, merges: string | string[]): Promise<void>;
|
|
22
|
+
/**
|
|
23
|
+
* Learn a byte-level BPE vocabulary + merges from a text corpus, then load
|
|
24
|
+
* them into this tokenizer. Makes the tokenizer self-contained (no external
|
|
25
|
+
* vocab file): train on your data, then `encode`/`decode` round-trips real
|
|
26
|
+
* text and frequent sequences compress to single tokens. The base vocabulary
|
|
27
|
+
* always covers all 256 byte symbols, so any input is representable.
|
|
28
|
+
*/
|
|
29
|
+
train(corpus: string | string[], opts?: {
|
|
30
|
+
numMerges?: number;
|
|
31
|
+
minPairFreq?: number;
|
|
32
|
+
}): void;
|
|
22
33
|
loadFromObjects(vocabObj: Record<string, number>, mergeArr: string[]): void;
|
|
23
34
|
encode(text: string, opts?: BPEEncodeOptions): number[];
|
|
24
35
|
decode(ids: number[]): string;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"bpe.d.ts","sourceRoot":"","sources":["../../src/tokenizer/bpe.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,MAAM,WAAW,gBAAgB;IAC/B,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,MAAM,CAAC,EAAE,OAAO,CAAC;CAClB;AAED,MAAM,MAAM,OAAO,GAAG,OAAO,GAAG,MAAM,CAAC;AA8BvC,qBAAa,YAAY;IACrB,KAAK,EAAE,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC3B,SAAS,EAAE,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC/B,MAAM,EAAE,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC5B,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;;IAef,IAAI,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,EAAE,MAAM,EAAE,MAAM,GAAG,MAAM,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC;IA6B5F,eAAe,CAAC,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,EAAE,QAAQ,EAAE,MAAM,EAAE,GAAG,IAAI;IAS3E,MAAM,CAAC,IAAI,EAAE,MAAM,EAAE,IAAI,GAAE,gBAAqB,GAAG,MAAM,EAAE;IA4B3D,MAAM,CAAC,GAAG,EAAE,MAAM,EAAE,GAAG,MAAM;IAgB7B,IAAI,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,EAAE;IA+B5B,aAAa,CAAC,GAAG,EAAE,MAAM,EAAE,EAAE,MAAM,EAAE,MAAM,EAAE,IAAI,GAAE,OAAiB,GAAG,MAAM,EAAE;IAO/E,IAAI,SAAS,IAAI,MAAM,CAA4B;CACtD"}
|
|
1
|
+
{"version":3,"file":"bpe.d.ts","sourceRoot":"","sources":["../../src/tokenizer/bpe.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,MAAM,WAAW,gBAAgB;IAC/B,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,MAAM,CAAC,EAAE,OAAO,CAAC;CAClB;AAED,MAAM,MAAM,OAAO,GAAG,OAAO,GAAG,MAAM,CAAC;AA8BvC,qBAAa,YAAY;IACrB,KAAK,EAAE,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC3B,SAAS,EAAE,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC/B,MAAM,EAAE,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC5B,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;;IAef,IAAI,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,EAAE,MAAM,EAAE,MAAM,GAAG,MAAM,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC;IA6B5F;;;;;;OAMG;IACH,KAAK,CAAC,MAAM,EAAE,MAAM,GAAG,MAAM,EAAE,EAAE,IAAI,GAAE;QAAE,SAAS,CAAC,EAAE,MAAM,CAAC;QAAC,WAAW,CAAC,EAAE,MAAM,CAAA;KAAO,GAAG,IAAI;IAsE/F,eAAe,CAAC,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,EAAE,QAAQ,EAAE,MAAM,EAAE,GAAG,IAAI;IAS3E,MAAM,CAAC,IAAI,EAAE,MAAM,EAAE,IAAI,GAAE,gBAAqB,GAAG,MAAM,EAAE;IA4B3D,MAAM,CAAC,GAAG,EAAE,MAAM,EAAE,GAAG,MAAM;IAgB7B,IAAI,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,EAAE;IA+B5B,aAAa,CAAC,GAAG,EAAE,MAAM,EAAE,EAAE,MAAM,EAAE,MAAM,EAAE,IAAI,GAAE,OAAiB,GAAG,MAAM,EAAE;IAO/E,IAAI,SAAS,IAAI,MAAM,CAA4B;CACtD"}
|
package/dist/tokenizer/bpe.js
CHANGED
|
@@ -76,6 +76,84 @@ export class BPETokenizer {
|
|
|
76
76
|
this.eosId = this.vocab.get(this.eosToken) ?? null;
|
|
77
77
|
this.padId = this.vocab.get(this.padToken) ?? null;
|
|
78
78
|
}
|
|
79
|
+
/**
|
|
80
|
+
* Learn a byte-level BPE vocabulary + merges from a text corpus, then load
|
|
81
|
+
* them into this tokenizer. Makes the tokenizer self-contained (no external
|
|
82
|
+
* vocab file): train on your data, then `encode`/`decode` round-trips real
|
|
83
|
+
* text and frequent sequences compress to single tokens. The base vocabulary
|
|
84
|
+
* always covers all 256 byte symbols, so any input is representable.
|
|
85
|
+
*/
|
|
86
|
+
train(corpus, opts = {}) {
|
|
87
|
+
const numMerges = opts.numMerges ?? 200;
|
|
88
|
+
const minPairFreq = opts.minPairFreq ?? 2;
|
|
89
|
+
const texts = Array.isArray(corpus) ? corpus : [corpus];
|
|
90
|
+
// Byte-encode each pre-token into a symbol sequence; count word frequencies.
|
|
91
|
+
const wordFreq = new Map();
|
|
92
|
+
for (const text of texts) {
|
|
93
|
+
for (const word of text.match(PRE_TOKENIZE_RE) ?? []) {
|
|
94
|
+
const bytes = new TextEncoder().encode(word);
|
|
95
|
+
const byteStr = Array.from(bytes).map((b) => BYTE_ENCODER.get(b) ?? '?').join('');
|
|
96
|
+
wordFreq.set(byteStr, (wordFreq.get(byteStr) ?? 0) + 1);
|
|
97
|
+
}
|
|
98
|
+
}
|
|
99
|
+
const words = [...wordFreq].map(([w, freq]) => ({ syms: [...w], freq }));
|
|
100
|
+
// Greedy BPE: repeatedly merge the most frequent adjacent symbol pair.
|
|
101
|
+
const merges = [];
|
|
102
|
+
for (let m = 0; m < numMerges; m++) {
|
|
103
|
+
const pairCounts = new Map();
|
|
104
|
+
for (const { syms, freq } of words) {
|
|
105
|
+
for (let i = 0; i < syms.length - 1; i++) {
|
|
106
|
+
const pair = syms[i] + ' ' + syms[i + 1];
|
|
107
|
+
pairCounts.set(pair, (pairCounts.get(pair) ?? 0) + freq);
|
|
108
|
+
}
|
|
109
|
+
}
|
|
110
|
+
let best = '';
|
|
111
|
+
let bestCount = 0;
|
|
112
|
+
for (const [pair, count] of pairCounts) {
|
|
113
|
+
if (count > bestCount) {
|
|
114
|
+
best = pair;
|
|
115
|
+
bestCount = count;
|
|
116
|
+
}
|
|
117
|
+
}
|
|
118
|
+
if (bestCount < minPairFreq)
|
|
119
|
+
break;
|
|
120
|
+
merges.push(best);
|
|
121
|
+
const sp = best.indexOf(' ');
|
|
122
|
+
const a = best.slice(0, sp);
|
|
123
|
+
const b = best.slice(sp + 1);
|
|
124
|
+
const merged = a + b;
|
|
125
|
+
for (const w of words) {
|
|
126
|
+
const out = [];
|
|
127
|
+
for (let i = 0; i < w.syms.length; i++) {
|
|
128
|
+
if (i < w.syms.length - 1 && w.syms[i] === a && w.syms[i + 1] === b) {
|
|
129
|
+
out.push(merged);
|
|
130
|
+
i++;
|
|
131
|
+
}
|
|
132
|
+
else {
|
|
133
|
+
out.push(w.syms[i]);
|
|
134
|
+
}
|
|
135
|
+
}
|
|
136
|
+
w.syms = out;
|
|
137
|
+
}
|
|
138
|
+
}
|
|
139
|
+
// Assemble the vocabulary: specials, every byte symbol (full coverage), then merges.
|
|
140
|
+
const vocabObj = {};
|
|
141
|
+
let id = 0;
|
|
142
|
+
for (const special of [this.unkToken, this.bosToken, this.eosToken, this.padToken]) {
|
|
143
|
+
if (!(special in vocabObj))
|
|
144
|
+
vocabObj[special] = id++;
|
|
145
|
+
}
|
|
146
|
+
for (const ch of BYTE_ENCODER.values()) {
|
|
147
|
+
if (!(ch in vocabObj))
|
|
148
|
+
vocabObj[ch] = id++;
|
|
149
|
+
}
|
|
150
|
+
for (const pair of merges) {
|
|
151
|
+
const merged = pair.split(' ').join('');
|
|
152
|
+
if (!(merged in vocabObj))
|
|
153
|
+
vocabObj[merged] = id++;
|
|
154
|
+
}
|
|
155
|
+
this.loadFromObjects(vocabObj, merges);
|
|
156
|
+
}
|
|
79
157
|
loadFromObjects(vocabObj, mergeArr) {
|
|
80
158
|
this.vocab = new Map(Object.entries(vocabObj).map(([k, v]) => [k, Number(v)]));
|
|
81
159
|
this.idToToken = new Map([...this.vocab].map(([k, v]) => [v, k]));
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"bpe.js","sourceRoot":"","sources":["../../src/tokenizer/bpe.ts"],"names":[],"mappings":"AAAA;;GAEG;AASH,SAAS,gBAAgB;IACrB,MAAM,GAAG,GAAG,IAAI,GAAG,EAAkB,CAAC;IACtC,MAAM,MAAM,GAAuB;QAC/B,CAAC,IAAI,EAAE,IAAI,CAAC;QACZ,CAAC,IAAI,EAAE,IAAI,CAAC;QACZ,CAAC,IAAI,EAAE,IAAI,CAAC;KACf,CAAC;IACF,IAAI,CAAC,GAAG,CAAC,CAAC;IACV,KAAK,MAAM,CAAC,EAAE,EAAE,EAAE,CAAC,IAAI,MAAM,EAAE,CAAC;QAC5B,KAAK,IAAI,CAAC,GAAG,EAAE,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,EAAE,EAAE,CAAC;YAC5B,GAAG,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC,CAAC;QACxC,CAAC;IACL,CAAC;IACD,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC;QAC3B,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC;YACd,GAAG,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM,CAAC,aAAa,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC;YAC1C,CAAC,EAAE,CAAC;QACR,CAAC;IACL,CAAC;IACD,OAAO,GAAG,CAAC;AACf,CAAC;AAED,MAAM,YAAY,GAAG,gBAAgB,EAAE,CAAC;AACxC,MAAM,YAAY,GAAG,IAAI,GAAG,CAAC,CAAC,GAAG,YAAY,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;AAExE,MAAM,eAAe,GACjB,sHAAsH,CAAC;AAE3H,MAAM,OAAO,YAAY;IACrB,KAAK,CAAsB;IAC3B,SAAS,CAAsB;IAC/B,MAAM,CAAsB;IAC5B,QAAQ,CAAS;IACjB,QAAQ,CAAS;IACjB,QAAQ,CAAS;IACjB,QAAQ,CAAS;IACjB,KAAK,CAAgB;IACrB,KAAK,CAAgB;IACrB,KAAK,CAAgB;IAErB;QACI,IAAI,CAAC,KAAK,GAAQ,IAAI,GAAG,EAAE,CAAC;QAC5B,IAAI,CAAC,SAAS,GAAI,IAAI,GAAG,EAAE,CAAC;QAC5B,IAAI,CAAC,MAAM,GAAO,IAAI,GAAG,EAAE,CAAC;QAC5B,IAAI,CAAC,QAAQ,GAAK,cAAc,CAAC;QACjC,IAAI,CAAC,QAAQ,GAAK,YAAY,CAAC;QAC/B,IAAI,CAAC,QAAQ,GAAK,eAAe,CAAC;QAClC,IAAI,CAAC,QAAQ,GAAK,OAAO,CAAC;QAC1B,IAAI,CAAC,KAAK,GAAQ,IAAI,CAAC;QACvB,IAAI,CAAC,KAAK,GAAQ,IAAI,CAAC;QACvB,IAAI,CAAC,KAAK,GAAQ,IAAI,CAAC;IAC3B,CAAC;IAED,KAAK,CAAC,IAAI,CAAC,KAAsC,EAAE,MAAyB;QACxE,IAAI,QAAgC,CAAC;QACrC,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;YAC5B,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,KAAK,CAAC,CAAC;YAC/B,QAAQ,GAAG,MAAM,GAAG,CAAC,IAAI,EAA4B,CAAC;QAC1D,CAAC;aAAM,CAAC;YACJ,QAAQ,GAAG,KAAK,CAAC;QACrB,CAAC;QACD,IAAI,CAAC,KAAK,GAAO,IAAI,GAAG,CAAC,MAAM,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;QACnF,IAAI,CAAC,SAAS,GAAG,IAAI,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;QAElE,IAAI,UAAoB,CAAC;QACzB,IAAI,OAAO,MAAM,KAAK,QAAQ,EAAE,CAAC;YAC7B,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,MAAM,CAAC,CAAC;YAChC,MAAM,GAAG,GAAG,MAAM,GAAG,CAAC,IAAI,EAAE,CAAC;YAC7B,UAAU,GAAG,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC;QACtE,CAAC;aAAM,CAAC;YACJ,UAAU,GAAG,MAAM,CAAC;QACxB,CAAC;QACD,IAAI,CAAC,MAAM,GAAG,IAAI,GAAG,EAAE,CAAC;QACxB,UAAU,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,IAAI,EAAE,EAAE;YAC9B,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,EAAE,EAAE,IAAI,CAAC,CAAC;QACvC,CAAC,CAAC,CAAC;QAEH,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,IAAI,CAAC;QACnD,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,IAAI,CAAC;QACnD,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,IAAI,CAAC;IACvD,CAAC;IAED,eAAe,CAAC,QAAgC,EAAE,QAAkB;QAChE,IAAI,CAAC,KAAK,GAAO,IAAI,GAAG,CAAC,MAAM,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;QACnF,IAAI,CAAC,SAAS,GAAG,IAAI,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;QAClE,IAAI,CAAC,MAAM,GAAM,IAAI,GAAG,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;QACzD,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,IAAI,CAAC;QACnD,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,IAAI,CAAC;QACnD,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,IAAI,CAAC;IACvD,CAAC;IAED,MAAM,CAAC,IAAY,EAAE,OAAyB,EAAE;QAC5C,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,eAAe,CAAC,IAAI,EAAE,CAAC;QAChD,MAAM,GAAG,GAAe,EAAE,CAAC;QAE3B,IAAI,IAAI,CAAC,MAAM,IAAI,IAAI,CAAC,KAAK,KAAK,IAAI;YAAE,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QAE7D,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;YACvB,MAAM,KAAK,GAAM,IAAI,WAAW,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;YAChD,MAAM,OAAO,GAAI,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,YAAY,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,GAAG,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;YACjF,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;YAErC,KAAK,MAAM,GAAG,IAAI,SAAS,EAAE,CAAC;gBAC1B,MAAM,EAAE,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;gBAC/B,IAAI,EAAE,KAAK,SAAS,EAAE,CAAC;oBACnB,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;gBACjB,CAAC;qBAAM,CAAC;oBACJ,KAAK,MAAM,EAAE,IAAI,GAAG,EAAE,CAAC;wBACnB,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;wBAC/B,IAAI,GAAG,KAAK,SAAS;4BAAE,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;oBACzC,CAAC;gBACL,CAAC;YACL,CAAC;QACL,CAAC;QAED,IAAI,IAAI,CAAC,MAAM,IAAI,IAAI,CAAC,KAAK,KAAK,IAAI;YAAE,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QAC7D,OAAO,GAAG,CAAC;IACf,CAAC;IAED,MAAM,CAAC,GAAa;QAChB,IAAI,OAAO,GAAG,EAAE,CAAC;QACjB,KAAK,MAAM,EAAE,IAAI,GAAG,EAAE,CAAC;YACnB,MAAM,GAAG,GAAG,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;YACnC,IAAI,GAAG,KAAK,SAAS;gBAAE,OAAO,IAAI,GAAG,CAAC;QAC1C,CAAC;QACD,MAAM,KAAK,GAAG,IAAI,UAAU,CACxB,CAAC,GAAG,OAAO,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,CAAC,YAAY,CAAC,GAAG,CAAC,EAAE,CAAC,IAAI,EAAE,CAAC,WAAW,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CACzE,CAAC;QACF,IAAI,CAAC;YACD,OAAO,IAAI,WAAW,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QAClD,CAAC;QAAC,MAAM,CAAC;YACL,OAAO,OAAO,CAAC;QACnB,CAAC;IACL,CAAC;IAED,IAAI,CAAC,IAAY;QACb,IAAI,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC;YAAE,OAAO,CAAC,IAAI,CAAC,CAAC;QAExC,IAAI,OAAO,GAAG,CAAC,GAAG,IAAI,CAAC,CAAC;QAExB,OAAO,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACxB,IAAI,QAAQ,GAAG,QAAQ,CAAC;YACxB,IAAI,OAAO,GAAI,CAAC,CAAC,CAAC;YAElB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;gBAC1C,MAAM,IAAI,GAAG,OAAO,CAAC,CAAC,CAAC,GAAG,GAAG,GAAG,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;gBAC/C,MAAM,IAAI,GAAG,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;gBACnC,IAAI,IAAI,KAAK,SAAS,IAAI,IAAI,GAAG,QAAQ,EAAE,CAAC;oBACxC,QAAQ,GAAG,IAAI,CAAC;oBAChB,OAAO,GAAI,CAAC,CAAC;gBACjB,CAAC;YACL,CAAC;YAED,IAAI,OAAO,KAAK,CAAC,CAAC;gBAAE,MAAM;YAE1B,MAAM,MAAM,GAAG,OAAO,CAAC,OAAO,CAAE,GAAG,OAAO,CAAC,OAAO,GAAG,CAAC,CAAE,CAAC;YACzD,OAAO,GAAG;gBACN,GAAG,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,OAAO,CAAC;gBAC5B,MAAM;gBACN,GAAG,OAAO,CAAC,KAAK,CAAC,OAAO,GAAG,CAAC,CAAC;aAChC,CAAC;QACN,CAAC;QAED,OAAO,OAAO,CAAC;IACnB,CAAC;IAED,aAAa,CAAC,GAAa,EAAE,MAAc,EAAE,OAAgB,OAAO;QAChE,IAAI,GAAG,CAAC,MAAM,IAAI,MAAM;YAAE,OAAO,GAAG,CAAC,KAAK,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC;QACtD,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,IAAI,CAAC,CAAC;QAC9B,MAAM,GAAG,GAAK,IAAI,KAAK,CAAS,MAAM,GAAG,GAAG,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QACjE,OAAO,IAAI,KAAK,OAAO,CAAC,CAAC,CAAC,CAAC,GAAG,GAAG,EAAE,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,GAAG,EAAE,GAAG,GAAG,CAAC,CAAC;IAClE,CAAC;IAED,IAAI,SAAS,KAAa,OAAO,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC;CACtD"}
|
|
1
|
+
{"version":3,"file":"bpe.js","sourceRoot":"","sources":["../../src/tokenizer/bpe.ts"],"names":[],"mappings":"AAAA;;GAEG;AASH,SAAS,gBAAgB;IACrB,MAAM,GAAG,GAAG,IAAI,GAAG,EAAkB,CAAC;IACtC,MAAM,MAAM,GAAuB;QAC/B,CAAC,IAAI,EAAE,IAAI,CAAC;QACZ,CAAC,IAAI,EAAE,IAAI,CAAC;QACZ,CAAC,IAAI,EAAE,IAAI,CAAC;KACf,CAAC;IACF,IAAI,CAAC,GAAG,CAAC,CAAC;IACV,KAAK,MAAM,CAAC,EAAE,EAAE,EAAE,CAAC,IAAI,MAAM,EAAE,CAAC;QAC5B,KAAK,IAAI,CAAC,GAAG,EAAE,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,EAAE,EAAE,CAAC;YAC5B,GAAG,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC,CAAC;QACxC,CAAC;IACL,CAAC;IACD,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC;QAC3B,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC;YACd,GAAG,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM,CAAC,aAAa,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC;YAC1C,CAAC,EAAE,CAAC;QACR,CAAC;IACL,CAAC;IACD,OAAO,GAAG,CAAC;AACf,CAAC;AAED,MAAM,YAAY,GAAG,gBAAgB,EAAE,CAAC;AACxC,MAAM,YAAY,GAAG,IAAI,GAAG,CAAC,CAAC,GAAG,YAAY,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;AAExE,MAAM,eAAe,GACjB,sHAAsH,CAAC;AAE3H,MAAM,OAAO,YAAY;IACrB,KAAK,CAAsB;IAC3B,SAAS,CAAsB;IAC/B,MAAM,CAAsB;IAC5B,QAAQ,CAAS;IACjB,QAAQ,CAAS;IACjB,QAAQ,CAAS;IACjB,QAAQ,CAAS;IACjB,KAAK,CAAgB;IACrB,KAAK,CAAgB;IACrB,KAAK,CAAgB;IAErB;QACI,IAAI,CAAC,KAAK,GAAQ,IAAI,GAAG,EAAE,CAAC;QAC5B,IAAI,CAAC,SAAS,GAAI,IAAI,GAAG,EAAE,CAAC;QAC5B,IAAI,CAAC,MAAM,GAAO,IAAI,GAAG,EAAE,CAAC;QAC5B,IAAI,CAAC,QAAQ,GAAK,cAAc,CAAC;QACjC,IAAI,CAAC,QAAQ,GAAK,YAAY,CAAC;QAC/B,IAAI,CAAC,QAAQ,GAAK,eAAe,CAAC;QAClC,IAAI,CAAC,QAAQ,GAAK,OAAO,CAAC;QAC1B,IAAI,CAAC,KAAK,GAAQ,IAAI,CAAC;QACvB,IAAI,CAAC,KAAK,GAAQ,IAAI,CAAC;QACvB,IAAI,CAAC,KAAK,GAAQ,IAAI,CAAC;IAC3B,CAAC;IAED,KAAK,CAAC,IAAI,CAAC,KAAsC,EAAE,MAAyB;QACxE,IAAI,QAAgC,CAAC;QACrC,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;YAC5B,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,KAAK,CAAC,CAAC;YAC/B,QAAQ,GAAG,MAAM,GAAG,CAAC,IAAI,EAA4B,CAAC;QAC1D,CAAC;aAAM,CAAC;YACJ,QAAQ,GAAG,KAAK,CAAC;QACrB,CAAC;QACD,IAAI,CAAC,KAAK,GAAO,IAAI,GAAG,CAAC,MAAM,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;QACnF,IAAI,CAAC,SAAS,GAAG,IAAI,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;QAElE,IAAI,UAAoB,CAAC;QACzB,IAAI,OAAO,MAAM,KAAK,QAAQ,EAAE,CAAC;YAC7B,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,MAAM,CAAC,CAAC;YAChC,MAAM,GAAG,GAAG,MAAM,GAAG,CAAC,IAAI,EAAE,CAAC;YAC7B,UAAU,GAAG,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC;QACtE,CAAC;aAAM,CAAC;YACJ,UAAU,GAAG,MAAM,CAAC;QACxB,CAAC;QACD,IAAI,CAAC,MAAM,GAAG,IAAI,GAAG,EAAE,CAAC;QACxB,UAAU,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,IAAI,EAAE,EAAE;YAC9B,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,EAAE,EAAE,IAAI,CAAC,CAAC;QACvC,CAAC,CAAC,CAAC;QAEH,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,IAAI,CAAC;QACnD,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,IAAI,CAAC;QACnD,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,IAAI,CAAC;IACvD,CAAC;IAED;;;;;;OAMG;IACH,KAAK,CAAC,MAAyB,EAAE,OAAqD,EAAE;QACpF,MAAM,SAAS,GAAG,IAAI,CAAC,SAAS,IAAI,GAAG,CAAC;QACxC,MAAM,WAAW,GAAG,IAAI,CAAC,WAAW,IAAI,CAAC,CAAC;QAC1C,MAAM,KAAK,GAAG,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC;QAExD,6EAA6E;QAC7E,MAAM,QAAQ,GAAG,IAAI,GAAG,EAAkB,CAAC;QAC3C,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;YACvB,KAAK,MAAM,IAAI,IAAI,IAAI,CAAC,KAAK,CAAC,eAAe,CAAC,IAAI,EAAE,EAAE,CAAC;gBACnD,MAAM,KAAK,GAAG,IAAI,WAAW,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;gBAC7C,MAAM,OAAO,GAAG,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,YAAY,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,GAAG,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;gBAClF,QAAQ,CAAC,GAAG,CAAC,OAAO,EAAE,CAAC,QAAQ,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;YAC5D,CAAC;QACL,CAAC;QACD,MAAM,KAAK,GAAG,CAAC,GAAG,QAAQ,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,IAAI,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC;QAEzE,uEAAuE;QACvE,MAAM,MAAM,GAAa,EAAE,CAAC;QAC5B,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,SAAS,EAAE,CAAC,EAAE,EAAE,CAAC;YACjC,MAAM,UAAU,GAAG,IAAI,GAAG,EAAkB,CAAC;YAC7C,KAAK,MAAM,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,KAAK,EAAE,CAAC;gBACjC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;oBACvC,MAAM,IAAI,GAAG,IAAI,CAAC,CAAC,CAAE,GAAG,GAAG,GAAG,IAAI,CAAC,CAAC,GAAG,CAAC,CAAE,CAAC;oBAC3C,UAAU,CAAC,GAAG,CAAC,IAAI,EAAE,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,GAAG,IAAI,CAAC,CAAC;gBAC7D,CAAC;YACL,CAAC;YACD,IAAI,IAAI,GAAG,EAAE,CAAC;YACd,IAAI,SAAS,GAAG,CAAC,CAAC;YAClB,KAAK,MAAM,CAAC,IAAI,EAAE,KAAK,CAAC,IAAI,UAAU,EAAE,CAAC;gBACrC,IAAI,KAAK,GAAG,SAAS,EAAE,CAAC;oBACpB,IAAI,GAAG,IAAI,CAAC;oBACZ,SAAS,GAAG,KAAK,CAAC;gBACtB,CAAC;YACL,CAAC;YACD,IAAI,SAAS,GAAG,WAAW;gBAAE,MAAM;YACnC,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YAClB,MAAM,EAAE,GAAG,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;YAC7B,MAAM,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;YAC5B,MAAM,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC;YAC7B,MAAM,MAAM,GAAG,CAAC,GAAG,CAAC,CAAC;YACrB,KAAK,MAAM,CAAC,IAAI,KAAK,EAAE,CAAC;gBACpB,MAAM,GAAG,GAAa,EAAE,CAAC;gBACzB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;oBACrC,IAAI,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,CAAC,EAAE,CAAC;wBAClE,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;wBACjB,CAAC,EAAE,CAAC;oBACR,CAAC;yBAAM,CAAC;wBACJ,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAE,CAAC,CAAC;oBACzB,CAAC;gBACL,CAAC;gBACD,CAAC,CAAC,IAAI,GAAG,GAAG,CAAC;YACjB,CAAC;QACL,CAAC;QAED,qFAAqF;QACrF,MAAM,QAAQ,GAA2B,EAAE,CAAC;QAC5C,IAAI,EAAE,GAAG,CAAC,CAAC;QACX,KAAK,MAAM,OAAO,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,IAAI,CAAC,QAAQ,EAAE,IAAI,CAAC,QAAQ,EAAE,IAAI,CAAC,QAAQ,CAAC,EAAE,CAAC;YACjF,IAAI,CAAC,CAAC,OAAO,IAAI,QAAQ,CAAC;gBAAE,QAAQ,CAAC,OAAO,CAAC,GAAG,EAAE,EAAE,CAAC;QACzD,CAAC;QACD,KAAK,MAAM,EAAE,IAAI,YAAY,CAAC,MAAM,EAAE,EAAE,CAAC;YACrC,IAAI,CAAC,CAAC,EAAE,IAAI,QAAQ,CAAC;gBAAE,QAAQ,CAAC,EAAE,CAAC,GAAG,EAAE,EAAE,CAAC;QAC/C,CAAC;QACD,KAAK,MAAM,IAAI,IAAI,MAAM,EAAE,CAAC;YACxB,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;YACxC,IAAI,CAAC,CAAC,MAAM,IAAI,QAAQ,CAAC;gBAAE,QAAQ,CAAC,MAAM,CAAC,GAAG,EAAE,EAAE,CAAC;QACvD,CAAC;QACD,IAAI,CAAC,eAAe,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAC;IAC3C,CAAC;IAED,eAAe,CAAC,QAAgC,EAAE,QAAkB;QAChE,IAAI,CAAC,KAAK,GAAO,IAAI,GAAG,CAAC,MAAM,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;QACnF,IAAI,CAAC,SAAS,GAAG,IAAI,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;QAClE,IAAI,CAAC,MAAM,GAAM,IAAI,GAAG,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;QACzD,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,IAAI,CAAC;QACnD,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,IAAI,CAAC;QACnD,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,IAAI,CAAC;IACvD,CAAC;IAED,MAAM,CAAC,IAAY,EAAE,OAAyB,EAAE;QAC5C,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,eAAe,CAAC,IAAI,EAAE,CAAC;QAChD,MAAM,GAAG,GAAe,EAAE,CAAC;QAE3B,IAAI,IAAI,CAAC,MAAM,IAAI,IAAI,CAAC,KAAK,KAAK,IAAI;YAAE,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QAE7D,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;YACvB,MAAM,KAAK,GAAM,IAAI,WAAW,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;YAChD,MAAM,OAAO,GAAI,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,YAAY,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,GAAG,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;YACjF,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;YAErC,KAAK,MAAM,GAAG,IAAI,SAAS,EAAE,CAAC;gBAC1B,MAAM,EAAE,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;gBAC/B,IAAI,EAAE,KAAK,SAAS,EAAE,CAAC;oBACnB,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;gBACjB,CAAC;qBAAM,CAAC;oBACJ,KAAK,MAAM,EAAE,IAAI,GAAG,EAAE,CAAC;wBACnB,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;wBAC/B,IAAI,GAAG,KAAK,SAAS;4BAAE,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;oBACzC,CAAC;gBACL,CAAC;YACL,CAAC;QACL,CAAC;QAED,IAAI,IAAI,CAAC,MAAM,IAAI,IAAI,CAAC,KAAK,KAAK,IAAI;YAAE,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QAC7D,OAAO,GAAG,CAAC;IACf,CAAC;IAED,MAAM,CAAC,GAAa;QAChB,IAAI,OAAO,GAAG,EAAE,CAAC;QACjB,KAAK,MAAM,EAAE,IAAI,GAAG,EAAE,CAAC;YACnB,MAAM,GAAG,GAAG,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;YACnC,IAAI,GAAG,KAAK,SAAS;gBAAE,OAAO,IAAI,GAAG,CAAC;QAC1C,CAAC;QACD,MAAM,KAAK,GAAG,IAAI,UAAU,CACxB,CAAC,GAAG,OAAO,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,CAAC,YAAY,CAAC,GAAG,CAAC,EAAE,CAAC,IAAI,EAAE,CAAC,WAAW,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CACzE,CAAC;QACF,IAAI,CAAC;YACD,OAAO,IAAI,WAAW,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QAClD,CAAC;QAAC,MAAM,CAAC;YACL,OAAO,OAAO,CAAC;QACnB,CAAC;IACL,CAAC;IAED,IAAI,CAAC,IAAY;QACb,IAAI,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC;YAAE,OAAO,CAAC,IAAI,CAAC,CAAC;QAExC,IAAI,OAAO,GAAG,CAAC,GAAG,IAAI,CAAC,CAAC;QAExB,OAAO,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACxB,IAAI,QAAQ,GAAG,QAAQ,CAAC;YACxB,IAAI,OAAO,GAAI,CAAC,CAAC,CAAC;YAElB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;gBAC1C,MAAM,IAAI,GAAG,OAAO,CAAC,CAAC,CAAC,GAAG,GAAG,GAAG,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;gBAC/C,MAAM,IAAI,GAAG,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;gBACnC,IAAI,IAAI,KAAK,SAAS,IAAI,IAAI,GAAG,QAAQ,EAAE,CAAC;oBACxC,QAAQ,GAAG,IAAI,CAAC;oBAChB,OAAO,GAAI,CAAC,CAAC;gBACjB,CAAC;YACL,CAAC;YAED,IAAI,OAAO,KAAK,CAAC,CAAC;gBAAE,MAAM;YAE1B,MAAM,MAAM,GAAG,OAAO,CAAC,OAAO,CAAE,GAAG,OAAO,CAAC,OAAO,GAAG,CAAC,CAAE,CAAC;YACzD,OAAO,GAAG;gBACN,GAAG,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,OAAO,CAAC;gBAC5B,MAAM;gBACN,GAAG,OAAO,CAAC,KAAK,CAAC,OAAO,GAAG,CAAC,CAAC;aAChC,CAAC;QACN,CAAC;QAED,OAAO,OAAO,CAAC;IACnB,CAAC;IAED,aAAa,CAAC,GAAa,EAAE,MAAc,EAAE,OAAgB,OAAO;QAChE,IAAI,GAAG,CAAC,MAAM,IAAI,MAAM;YAAE,OAAO,GAAG,CAAC,KAAK,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC;QACtD,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,IAAI,CAAC,CAAC;QAC9B,MAAM,GAAG,GAAK,IAAI,KAAK,CAAS,MAAM,GAAG,GAAG,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QACjE,OAAO,IAAI,KAAK,OAAO,CAAC,CAAC,CAAC,CAAC,GAAG,GAAG,EAAE,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,GAAG,EAAE,GAAG,GAAG,CAAC,CAAC;IAClE,CAAC;IAED,IAAI,SAAS,KAAa,OAAO,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC;CACtD"}
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@seanhogg/builderforce-memory-engine",
|
|
3
|
-
"version": "2026.6.
|
|
3
|
+
"version": "2026.6.30",
|
|
4
4
|
"description": "BuilderForce Agent Memory — engine layer. TypeScript/WGSL Mamba SSM kernels, model blocks, autograd, training, and BPE tokenizer (zero runtime deps, WebGPU-native).",
|
|
5
5
|
"main": "dist/index.js",
|
|
6
6
|
"module": "dist/index.js",
|
|
@@ -60,7 +60,8 @@
|
|
|
60
60
|
"eslint": "^8.57.0",
|
|
61
61
|
"jest": "^29.7.0",
|
|
62
62
|
"ts-jest": "^29.2.0",
|
|
63
|
-
"typescript": "^
|
|
63
|
+
"typescript": "^6.0.3",
|
|
64
|
+
"@typescript/native-preview": "7.0.0-dev.20260627.2"
|
|
64
65
|
},
|
|
65
66
|
"jest": {
|
|
66
67
|
"preset": "ts-jest/presets/default-esm",
|
|
@@ -94,6 +95,7 @@
|
|
|
94
95
|
"build": "tsc",
|
|
95
96
|
"serve": "node serve.js",
|
|
96
97
|
"test": "node --experimental-vm-modules node_modules/jest/bin/jest.js",
|
|
97
|
-
"lint": "eslint src/ tests/"
|
|
98
|
+
"lint": "eslint src/ tests/",
|
|
99
|
+
"typecheck:native": "tsgo --noEmit"
|
|
98
100
|
}
|
|
99
101
|
}
|
package/src/lm/evermind_lm.ts
CHANGED
|
@@ -28,20 +28,20 @@ import { SeededRng } from "../utils/rng.js";
|
|
|
28
28
|
import { quantizeFp16, dequantizeFp16 } from "../utils/quantization.js";
|
|
29
29
|
|
|
30
30
|
export interface EvermindLMConfig {
|
|
31
|
-
/** Vocabulary size. */
|
|
31
|
+
/** Vocabulary size (the only required field; everything else has a default). */
|
|
32
32
|
vocabSize: number;
|
|
33
33
|
/** Model (channel) dimension. Default 64. */
|
|
34
|
-
dModel
|
|
34
|
+
dModel?: number;
|
|
35
35
|
/** Number of (conv + MoE) blocks. Default 2. */
|
|
36
|
-
numLayers
|
|
36
|
+
numLayers?: number;
|
|
37
37
|
/** Causal conv kernel width. Default 3. */
|
|
38
|
-
convKernel
|
|
38
|
+
convKernel?: number;
|
|
39
39
|
/** Hidden width of each MoE expert FFN. Default 2·dModel. */
|
|
40
|
-
hiddenDim
|
|
40
|
+
hiddenDim?: number;
|
|
41
41
|
/** Routed experts per MoE layer. Default 4. */
|
|
42
|
-
numExperts
|
|
42
|
+
numExperts?: number;
|
|
43
43
|
/** Experts activated per token. Default 2. */
|
|
44
|
-
topK
|
|
44
|
+
topK?: number;
|
|
45
45
|
/** Deterministic init seed. */
|
|
46
46
|
seed?: number;
|
|
47
47
|
}
|
package/src/tokenizer/bpe.ts
CHANGED
|
@@ -91,6 +91,83 @@ export class BPETokenizer {
|
|
|
91
91
|
this.padId = this.vocab.get(this.padToken) ?? null;
|
|
92
92
|
}
|
|
93
93
|
|
|
94
|
+
/**
|
|
95
|
+
* Learn a byte-level BPE vocabulary + merges from a text corpus, then load
|
|
96
|
+
* them into this tokenizer. Makes the tokenizer self-contained (no external
|
|
97
|
+
* vocab file): train on your data, then `encode`/`decode` round-trips real
|
|
98
|
+
* text and frequent sequences compress to single tokens. The base vocabulary
|
|
99
|
+
* always covers all 256 byte symbols, so any input is representable.
|
|
100
|
+
*/
|
|
101
|
+
train(corpus: string | string[], opts: { numMerges?: number; minPairFreq?: number } = {}): void {
|
|
102
|
+
const numMerges = opts.numMerges ?? 200;
|
|
103
|
+
const minPairFreq = opts.minPairFreq ?? 2;
|
|
104
|
+
const texts = Array.isArray(corpus) ? corpus : [corpus];
|
|
105
|
+
|
|
106
|
+
// Byte-encode each pre-token into a symbol sequence; count word frequencies.
|
|
107
|
+
const wordFreq = new Map<string, number>();
|
|
108
|
+
for (const text of texts) {
|
|
109
|
+
for (const word of text.match(PRE_TOKENIZE_RE) ?? []) {
|
|
110
|
+
const bytes = new TextEncoder().encode(word);
|
|
111
|
+
const byteStr = Array.from(bytes).map((b) => BYTE_ENCODER.get(b) ?? '?').join('');
|
|
112
|
+
wordFreq.set(byteStr, (wordFreq.get(byteStr) ?? 0) + 1);
|
|
113
|
+
}
|
|
114
|
+
}
|
|
115
|
+
const words = [...wordFreq].map(([w, freq]) => ({ syms: [...w], freq }));
|
|
116
|
+
|
|
117
|
+
// Greedy BPE: repeatedly merge the most frequent adjacent symbol pair.
|
|
118
|
+
const merges: string[] = [];
|
|
119
|
+
for (let m = 0; m < numMerges; m++) {
|
|
120
|
+
const pairCounts = new Map<string, number>();
|
|
121
|
+
for (const { syms, freq } of words) {
|
|
122
|
+
for (let i = 0; i < syms.length - 1; i++) {
|
|
123
|
+
const pair = syms[i]! + ' ' + syms[i + 1]!;
|
|
124
|
+
pairCounts.set(pair, (pairCounts.get(pair) ?? 0) + freq);
|
|
125
|
+
}
|
|
126
|
+
}
|
|
127
|
+
let best = '';
|
|
128
|
+
let bestCount = 0;
|
|
129
|
+
for (const [pair, count] of pairCounts) {
|
|
130
|
+
if (count > bestCount) {
|
|
131
|
+
best = pair;
|
|
132
|
+
bestCount = count;
|
|
133
|
+
}
|
|
134
|
+
}
|
|
135
|
+
if (bestCount < minPairFreq) break;
|
|
136
|
+
merges.push(best);
|
|
137
|
+
const sp = best.indexOf(' ');
|
|
138
|
+
const a = best.slice(0, sp);
|
|
139
|
+
const b = best.slice(sp + 1);
|
|
140
|
+
const merged = a + b;
|
|
141
|
+
for (const w of words) {
|
|
142
|
+
const out: string[] = [];
|
|
143
|
+
for (let i = 0; i < w.syms.length; i++) {
|
|
144
|
+
if (i < w.syms.length - 1 && w.syms[i] === a && w.syms[i + 1] === b) {
|
|
145
|
+
out.push(merged);
|
|
146
|
+
i++;
|
|
147
|
+
} else {
|
|
148
|
+
out.push(w.syms[i]!);
|
|
149
|
+
}
|
|
150
|
+
}
|
|
151
|
+
w.syms = out;
|
|
152
|
+
}
|
|
153
|
+
}
|
|
154
|
+
|
|
155
|
+
// Assemble the vocabulary: specials, every byte symbol (full coverage), then merges.
|
|
156
|
+
const vocabObj: Record<string, number> = {};
|
|
157
|
+
let id = 0;
|
|
158
|
+
for (const special of [this.unkToken, this.bosToken, this.eosToken, this.padToken]) {
|
|
159
|
+
if (!(special in vocabObj)) vocabObj[special] = id++;
|
|
160
|
+
}
|
|
161
|
+
for (const ch of BYTE_ENCODER.values()) {
|
|
162
|
+
if (!(ch in vocabObj)) vocabObj[ch] = id++;
|
|
163
|
+
}
|
|
164
|
+
for (const pair of merges) {
|
|
165
|
+
const merged = pair.split(' ').join('');
|
|
166
|
+
if (!(merged in vocabObj)) vocabObj[merged] = id++;
|
|
167
|
+
}
|
|
168
|
+
this.loadFromObjects(vocabObj, merges);
|
|
169
|
+
}
|
|
170
|
+
|
|
94
171
|
loadFromObjects(vocabObj: Record<string, number>, mergeArr: string[]): void {
|
|
95
172
|
this.vocab = new Map(Object.entries(vocabObj).map(([k, v]) => [k, Number(v)]));
|
|
96
173
|
this.idToToken = new Map([...this.vocab].map(([k, v]) => [v, k]));
|