@genai-fi/nanogpt 0.10.2 → 0.11.0

package/dist/{RealDiv-zz7FpkKX.js → RealDiv-Ds-jvL09.js}

@@ -1,12 +1,10 @@
-import "./index-D6Q1lPZO.js";
-import { r as $ } from "./Reshape-CHdUjC72.js";
-import { _ as T, g as E, y as B, $ as F } from "./tensor_util-DfwaWayG.js";
-import { G as _, e as G, p as O, s as V } from "./tensor-CzmOBsdf.js";
-import { a as A, b as k, d as C, c as N, e as R } from "./axis_util-BsIr9ZNu.js";
-import { t as K, m as U } from "./shared-CHhxz-O5.js";
-import { c as W } from "./backend_util-B1XRLuq9.js";
-import { f as y } from "./gpgpu_math-DsCcikas.js";
-import { g as j, b as L } from "./kernel_funcs_utils-CWfOAPGO.js";
+import { aG as T, ab as E, af as O, V, aS as B, Q as F, am as G, aT as K } from "./index-D0RBWjq8.js";
+import { r as $ } from "./Reshape-Ct266DEk.js";
+import { a as A, b as k, d as C, c as N, e as R } from "./axis_util-DofAuy0p.js";
+import { t as U, m as W } from "./shared-B7USJZgw.js";
+import { c as _ } from "./backend_util-C7NWHpv7.js";
+import { f as y } from "./gpgpu_math-CJcbnKPC.js";
+import { g as j, b as L } from "./kernel_funcs_utils-CSaumNDs.js";
 class w {
   constructor(s, e) {
     this.variableNames = ["x"];
@@ -16,7 +14,7 @@ class w {
     let o = "sumValue += dot(values, ones);";
     if (e != null) {
       const p = 1 / e;
-      o = `sumValue += dot(values * ${_(p) ? p.toPrecision(2) : p}, ones);`;
+      o = `sumValue += dot(values * ${T(p) ? p.toPrecision(2) : p}, ones);`;
     }
     let u = "";
     l % t > 0 && (u = `
@@ -186,7 +184,7 @@ class X {
 function q(a) {
   const s = [];
   for (; s.length === 0 || s[s.length - 1].outSize !== 1; ) {
-    const e = s.length ? s[s.length - 1].outSize : a[1], t = W(e);
+    const e = s.length ? s[s.length - 1].outSize : a[1], t = _(e);
     s.push({
       inSize: e,
       windowSize: t,
@@ -205,14 +203,14 @@ function P(a, s, e, t) {
   }
   return l;
 }
-class Y {
+class Q {
   constructor(s, e) {
     this.variableNames = ["A"];
     const t = new Array(s.length);
     for (let r = 0; r < t.length; r++)
       t[r] = s[e[r]];
     this.outputShape = t, this.rank = t.length;
-    const n = y(this.rank), l = H(e);
+    const n = y(this.rank), l = Y(e);
     this.userCode = `
     void main() {
       ${n} resRC = getOutputCoords();
@@ -221,7 +219,7 @@ class Y {
     `;
   }
 }
-function H(a) {
+function Y(a) {
   const s = a.length;
   if (s > 6)
     throw Error(`Transpose for rank ${s} is not yet supported`);
@@ -230,7 +228,7 @@ function H(a) {
     t[a[n]] = e[n];
   return t.join();
 }
-class J {
+class H {
   constructor(s, e) {
     this.variableNames = ["A"], this.packedInputs = !0, this.packedOutput = !0;
     const t = new Array(s.length);
@@ -263,10 +261,10 @@ class J {
   }
 }
 function D(a, s, e) {
-  const t = G().getBool("WEBGL_PACK_ARRAY_OPERATIONS") ? new J(a.shape, s) : new Y(a.shape, s);
+  const t = E().getBool("WEBGL_PACK_ARRAY_OPERATIONS") ? new H(a.shape, s) : new Q(a.shape, s);
   return e.runWebGLProgram(t, [a], a.dtype);
 }
-function Q(a, s, e, t) {
+function J(a, s, e, t) {
   const n = s, l = a.shape.length, r = O(n, a.shape);
   let i = r;
   const c = A(i, l), o = c != null;
@@ -275,15 +273,15 @@ function Q(a, s, e, t) {
   const [p, h] = N(u.shape, i);
   let d = p;
   e && (d = R(p, r));
-  const f = V(h), g = V(a.shape) / f, x = $({ inputs: { x: u }, attrs: { shape: [g, f] }, backend: t }), b = T(a.dtype), I = P(x, b, "sum", t), m = $({ inputs: { x: I }, attrs: { shape: d }, backend: t });
+  const f = V(h), g = V(a.shape) / f, x = $({ inputs: { x: u }, attrs: { shape: [g, f] }, backend: t }), b = B(a.dtype), I = P(x, b, "sum", t), m = $({ inputs: { x: I }, attrs: { shape: d }, backend: t });
   return t.disposeIntermediateTensorInfo(x), t.disposeIntermediateTensorInfo(I), o && t.disposeIntermediateTensorInfo(u), m;
 }
 function Z(a) {
   const { inputs: s, backend: e, attrs: t } = a, { x: n } = s, { axis: l, keepDims: r } = t;
-  return Q(n, l, r, e);
+  return J(n, l, r, e);
 }
-const fe = {
-  kernelName: E,
+const pe = {
+  kernelName: F,
   backendName: "webgl",
   kernelFunc: Z
 };
@@ -301,7 +299,7 @@ function te(a) {
       const I = e.texData.get(d.dataId).values, m = new Array(i);
       for (let v = 0; v < m.length; v++)
         m[v] = n.shape[u[v]];
-      const z = K(I, n.shape, n.dtype, u, m);
+      const z = U(I, n.shape, n.dtype, u, m);
       d = e.makeTensorInfo(m, n.dtype);
       const M = e.texData.get(d.dataId);
       M.values = z;
@@ -315,7 +313,7 @@ function te(a) {
   r && (g = R(f, c));
   let x;
   if (h) {
-    const I = e.texData.get(d.dataId).values, m = U(I, V(S), g, n.dtype);
+    const I = e.texData.get(d.dataId).values, m = W(I, V(S), g, n.dtype);
     x = e.makeTensorInfo(g, n.dtype);
     const z = e.texData.get(x.dataId);
     z.values = m;
@@ -323,8 +321,8 @@ function te(a) {
     x = ee(d, S, g, e);
   return p && e.disposeIntermediateTensorInfo(d), x;
 }
-const me = {
-  kernelName: B,
+const he = {
+  kernelName: G,
   backendName: "webgl",
   kernelFunc: te
 };
@@ -350,16 +348,16 @@ return a / b;`, se = `
   }
 
   return result;
-`, ne = L({ opSnippet: ae, packedOpSnippet: se, checkOutOfBounds: !0 }), xe = {
-  kernelName: F,
+`, ne = L({ opSnippet: ae, packedOpSnippet: se, checkOutOfBounds: !0 }), fe = {
+  kernelName: K,
   backendName: "webgl",
   kernelFunc: ne
 };
 export {
   P as a,
-  me as b,
-  xe as c,
-  fe as d,
+  he as b,
+  fe as c,
+  pe as d,
   te as m,
   ne as r,
   Z as s,

package/dist/Reshape-Cd6e-Otn.js

@@ -0,0 +1,14 @@
+import { V as h, ah as d, y as c, R as m } from "./index-D0RBWjq8.js";
+function i(n) {
+  const { inputs: p, attrs: o } = n, { x: e } = p, { shape: r } = o, a = h(e.shape), s = d(r, a), t = h(s);
+  return c(a === t, () => `The new shape (${s}) has ${t} elements and the old shape (${e.shape}) has ${a} elements. The new shape and old shape must have the same number of elements.`), n.backend.incRef(e.dataId), { dataId: e.dataId, shape: s, dtype: e.dtype };
+}
+const u = {
+  kernelName: m,
+  backendName: "webgpu",
+  kernelFunc: i
+};
+export {
+  u as a,
+  i as r
+};

package/dist/{Reshape-CHdUjC72.js → Reshape-Ct266DEk.js}

@@ -1,10 +1,8 @@
-import "./index-D6Q1lPZO.js";
-import { u as C, g as f, a as R, b as g, c as I, d as c, e as u, i as m } from "./gpgpu_math-DsCcikas.js";
-import { b as x } from "./tensor_util-DfwaWayG.js";
-import { s as l, n as F, a as $ } from "./tensor-CzmOBsdf.js";
+import { R as C, V as c, ah as R, y as f } from "./index-D0RBWjq8.js";
+import { u as g, g as I, a as x, b as F, c as $, d as u, e as m, i as l } from "./gpgpu_math-CJcbnKPC.js";
 class S {
   constructor(t, i) {
-    this.variableNames = ["A"], this.packedInputs = !0, this.packedOutput = !0, this.customUniforms = [{ name: "inputShape", type: "ivec3" }], this.outputShape = t, this.enableShapeUniforms = C(this.outputShape.length);
+    this.variableNames = ["A"], this.packedInputs = !0, this.packedOutput = !0, this.customUniforms = [{ name: "inputShape", type: "ivec3" }], this.outputShape = t, this.enableShapeUniforms = g(this.outputShape.length);
     let a = "";
     for (let e = 0; e < 4; e++) {
       let o = "thisRC = rc;";
@@ -22,8 +20,8 @@ class S {
       `;
     }
     this.userCode = `
-      ${b(i, this.enableShapeUniforms)}
-      ${this.enableShapeUniforms ? f() : R(t)}
+      ${v(i, this.enableShapeUniforms)}
+      ${this.enableShapeUniforms ? I() : x(t)}
 
       void main() {
         ivec3 rc = getOutputCoords();
@@ -41,41 +39,41 @@ class S {
     `;
   }
 }
-function b(s, t) {
+function v(s, t) {
   return `
     ivec3 inputCoordsFromReshapedOutCoords(int index) {
-      ${t ? g(["r", "c", "d"], "inputShape") : I(["r", "c", "d"], s)}
+      ${t ? F(["r", "c", "d"], "inputShape") : $(["r", "c", "d"], s)}
       return ivec3(r, c, d);
     }
   `;
 }
-function v(s, t, i) {
+function y(s, t, i) {
   const a = [
-    c(s.shape),
-    ...u(s.shape)
+    u(s.shape),
+    ...m(s.shape)
   ], e = {
     dtype: s.dtype,
     shape: a,
     dataId: s.dataId
   }, o = [
-    c(t),
-    ...u(t)
+    u(t),
+    ...m(t)
   ], r = new S(o, a), p = !0, n = [a], h = i.runWebGLProgram(r, [e], s.dtype, n, p);
   return { dataId: h.dataId, shape: t, dtype: h.dtype };
 }
-function y(s) {
-  const { inputs: t, backend: i, attrs: a } = s, { x: e } = t, { shape: o } = a, r = i, p = l(e.shape), n = F(o, p), h = l(n);
-  $(p === h, () => `The new shape (${n}) has ${h} elements and the old shape (${e.shape}) has ${p} elements. The new shape and old shape must have the same number of elements.`);
+function b(s) {
+  const { inputs: t, backend: i, attrs: a } = s, { x: e } = t, { shape: o } = a, r = i, p = c(e.shape), n = R(o, p), h = c(n);
+  f(p === h, () => `The new shape (${n}) has ${h} elements and the old shape (${e.shape}) has ${p} elements. The new shape and old shape must have the same number of elements.`);
   const d = r.texData.get(e.dataId);
-  return d.isPacked && !m(e.shape, n) && !(d.texture !== null && m(d.shape, n)) ? v(e, n, r) : (r.incRef(e.dataId), { dataId: e.dataId, shape: n, dtype: e.dtype });
+  return d.isPacked && !l(e.shape, n) && !(d.texture !== null && l(d.shape, n)) ? y(e, n, r) : (r.incRef(e.dataId), { dataId: e.dataId, shape: n, dtype: e.dtype });
 }
-const O = {
-  kernelName: x,
+const U = {
+  kernelName: C,
   backendName: "webgl",
-  kernelFunc: y
+  kernelFunc: b
 };
 export {
   S as R,
-  O as a,
-  y as r
+  U as a,
+  b as r
 };

package/dist/TeachableLLM.d.ts

@@ -1,5 +1,5 @@
 import { GPTConfig } from './models/config';
-import { ITokeniser } from './tokeniser/type';
+import { Conversation, ITokeniser } from './tokeniser/type';
 import { SaveOptions } from './loader/save';
 import { default as Generator, IGenerateOptions } from './Generator';
 import { default as Trainer, ITrainerOptions } from './Trainer';
@@ -41,10 +41,11 @@ export default class TeachableLLM {
     set enableProfiler(value: boolean);
     getNumParams(): number;
     trainer(): Trainer;
-    train(text: string[], options?: ITrainerOptions): Promise<void>;
+    train(text: Conversation[][], options?: ITrainerOptions): Promise<void>;
     trainTokeniser(text: string[]): Promise<number>;
     generator(): Generator;
-    generateText(prompt?: string, options?: IGenerateOptions): Promise<string>;
+    generateText(prompt: Conversation[], options?: IGenerateOptions): Promise<Conversation[]>;
+    generateText(options?: IGenerateOptions): Promise<Conversation[]>;
     dispose(): void;
     on(event: 'status', listener: (status: TeachableLLMStatus) => void): void;
     on(event: 'error', listener: (error: Error) => void): void;

package/dist/TeachableLLM.js

@@ -2,28 +2,27 @@ import { defaultConfig as d } from "./models/config.js";
 import { saveModel as l } from "./loader/save.js";
 import { loadModel as _ } from "./loader/load.js";
 import u from "./Generator.js";
-import p from "./Trainer.js";
-import { E as f } from "./index-DvYrXKkX.js";
+import f from "./Trainer.js";
+import { E as p } from "./index-DvYrXKkX.js";
 import { dummyPassTrainAsync as m } from "./utilities/dummy.js";
-import "./utilities/packed.js";
-import "./index-D6Q1lPZO.js";
+import "./index-D0RBWjq8.js";
+import "./random_width-UGQn4OWb.js";
+import "./zeros_like-BAz3iKru.js";
+import "./index-Cp39cXWe.js";
+import "./dataset-DcjWqUVQ.js";
 import "./ops/cpu/attentionMask.js";
 import "./ops/webgl/attentionMask.js";
 import "./ops/grads/attentionMask.js";
-import "./random_width-BVV9HveY.js";
-import "./register_all_kernels-nvj2k7OC.js";
-import "./index-Cp39cXWe.js";
-import "./dataset-D2P7rHAw.js";
 import "./ops/cpu/rope.js";
 import "./ops/webgl/rope.js";
-import "./rope-s4W2XO9B.js";
+import "./rope-BmZmp9uP.js";
 import "./ops/cpu/appendCache.js";
 import "./ops/webgl/appendCache.js";
 import "./ops/grads/softmax16.js";
-import "./matMul16-fEAJ4smh.js";
+import "./matMul16-cDxwemKj.js";
 import "./ops/webgl/matMul16.js";
 import "./ops/cpu/matMul16.js";
-import "./pack16-CmVZs6af.js";
+import "./pack16-k4jq6aMX.js";
 import "./ops/transpose16.js";
 import "./ops/reshape16.js";
 import "./ops/cpu/qkv.js";
@@ -42,11 +41,11 @@ import g from "./tokeniser/bpe.js";
 import "./papaparse.min-C0cScC2i.js";
 import "./jszip.min-Bz5-11Bk.js";
 import "./ops/cpu/matMulGelu.js";
-import "./ops/webgl/matMulGelu.js";
+import "./matMulGelu-B2s_80-H.js";
 import "./ops/grads/matMulGelu.js";
 import "./ops/cpu/gelu.js";
 import "./ops/webgl/gelu.js";
-import "./gelu-Bmhopi0J.js";
+import "./gelu-DqTbCx5x.js";
 import "./ops/webgl/log.js";
 import "./ops/cpu/adamMoments.js";
 import "./ops/webgl/adamMoments.js";
@@ -57,7 +56,7 @@ import "./checks/normRMSGrad.js";
 import k from "./utilities/profile.js";
 import w from "./models/factory.js";
 class a {
-  ee = new f();
+  ee = new p();
   _config;
   _model;
   _tokeniser;
@@ -156,7 +155,7 @@ class a {
   trainer() {
     if (!this._model || !this._tokeniser)
       throw new Error("model_or_tokeniser_not_initialized.");
-    const t = new p(this._model, this._tokeniser);
+    const t = new f(this._model, this._tokeniser);
     return t.on("start", () => this.setStatus("training")), t.on("stop", () => this.setStatus("ready")), t.on("log", async (e, r) => {
       const o = this.ee.listeners("trainStep");
       for (const s of o)
@@ -184,7 +183,7 @@ class a {
     }), t;
   }
   generateText(t, e) {
-    return this.generator().generate(t, e);
+    return Array.isArray(t) ? this.generator().generate(t, e) : this.generator().generate([], e);
   }
   dispose() {
     this._model?.dispose();

package/dist/Trainer.d.ts

@@ -1,4 +1,4 @@
-import { ITokeniser } from './tokeniser/type';
+import { Conversation, ITokeniser } from './tokeniser/type';
 import { default as EE } from 'eventemitter3';
 import { TrainingLogEntry, TrainingProgress } from './training/Trainer';
 import { default as Model, ModelForwardAttributes } from './models/model';
@@ -30,7 +30,7 @@ export default class Trainer extends EE<'start' | 'stop' | 'log'> {
     constructor(model: Model<ModelForwardAttributes>, tokeniser: ITokeniser);
     stop(): void;
     reset(): void;
-    prepare(text: string[], options?: ITrainerOptions): Promise<void>;
+    prepare(text: Conversation[][], options?: ITrainerOptions): Promise<void>;
     train(options?: ITrainerOptions): Promise<void>;
     step(options?: ITrainerOptions): Promise<void>;
     getLog(): TrainingLogEntry[];

package/dist/Trainer.js

@@ -1,6 +1,6 @@
-import { E as l } from "./index-DvYrXKkX.js";
-import h from "./training/FullTrainer.js";
-class m extends l {
+import { E as o } from "./index-DvYrXKkX.js";
+import d from "./training/FullTrainer.js";
+class g extends o {
   trainer;
   hasTrained = !1;
   trainDataset;
@@ -9,7 +9,7 @@ class m extends l {
   log = [];
   progress = null;
   constructor(t, e) {
-    super(), this.trainer = new h(t, e, 1e-3);
+    super(), this.trainer = new d(t, e, 1e-3);
   }
   stop() {
     this.trainer.stop();
@@ -22,7 +22,7 @@ class m extends l {
       t,
       e?.batchSize || 32,
       e?.validationSplit || 0.1
-    ), i = t.reduce((r, n) => r + n.length, 0) * (1 - (e?.validationSplit || 0));
+    ), i = t.reduce((r, n) => r + n.reduce((l, h) => l + h.content.length, 0), 0) * (1 - (e?.validationSplit || 0));
     this.trainDataset = a, this.validationDataset = s, this.totalSamples = i;
   }
   async train(t) {
@@ -91,5 +91,5 @@ class m extends l {
   }
 }
 export {
-  m as default
+  g as default
 };

package/dist/{axis_util-BsIr9ZNu.js → axis_util-DofAuy0p.js}

@@ -1,4 +1,4 @@
-import { a as c } from "./tensor-CzmOBsdf.js";
+import { y as c } from "./index-D0RBWjq8.js";
 function i(e, n) {
   for (let t = 0; t < e.length; ++t)
     if (e[e.length - t - 1] !== n - 1 - t)

package/dist/backend.js

@@ -1,9 +1,9 @@
-import { g as o, s as e, r as s } from "./index-D6Q1lPZO.js";
+import { g as o, s as e, r as s } from "./index-D0RBWjq8.js";
 async function c(t, a) {
   if (o() !== t) {
     if (t === "webgpu") {
       const { registerWebGPUBackend: i } = await import("./patches/webgpu_base.js");
-      i(a), await import("./index-DRyE072i.js"), await import("./ops/webgpu/index.js");
+      i(a), await import("./index-Dj5TkmPY.js"), await import("./ops/webgpu/index.js");
     }
     await e(t), await s(), console.log(`Backend set to ${t}`);
   }