@genai-fi/nanogpt 0.2.12 → 0.3.1

package/dist/layers/TiedEmbedding.js

@@ -1,438 +1,36 @@
-import { o as h, d as i, E as o, K as X, N as Y, O as Z, Q as J, T as ee, U as te, V as se, W as ne, X as re, Y as ue, l as L, I as ae, Z as A, a as ie, _ as oe, D as le, f as q, v as C, $ as P, H as U, a0 as H } from "../index-YPKosni4.js";
-import { r as f } from "../reshape-DmnmKT6r.js";
-import { s as ce } from "../sum-D7fu15XL.js";
-import { m } from "../mat_mul-Bu7bhLms.js";
-import { c as pe } from "../complex-CJ-qCcLB.js";
-/**
- * @license
- * Copyright 2018 Google LLC. All Rights Reserved.
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- * =============================================================================
- */
-function he(t) {
-  const s = { x: i(t, "x", "sigmoid", "float32") };
-  return o.runKernel(X, s);
-}
-const fe = /* @__PURE__ */ h({ sigmoid_: he });
-/**
- * @license
- * Copyright 2020 Google LLC. All Rights Reserved.
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- * =============================================================================
- */
-function de(t) {
-  const s = { x: i(t, "x", "elu", "float32") };
-  return o.runKernel(Y, s);
-}
-const me = /* @__PURE__ */ h({ elu_: de });
-/**
- * @license
- * Copyright 2020 Google LLC. All Rights Reserved.
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- * =============================================================================
- */
-function ge(t) {
-  const s = { input: i(t, "input", "imag") };
-  return o.runKernel(Z, s);
-}
-const $e = /* @__PURE__ */ h({ imag_: ge });
-/**
- * @license
- * Copyright 2020 Google LLC. All Rights Reserved.
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- * =============================================================================
- */
-function xe(t, e = 0.2) {
-  const n = { x: i(t, "x", "leakyRelu") }, r = { alpha: e };
-  return o.runKernel(J, n, r);
-}
-const ke = /* @__PURE__ */ h({ leakyRelu_: xe });
-/**
- * @license
- * Copyright 2018 Google LLC. All Rights Reserved.
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- * =============================================================================
- */
-function De(t) {
-  const s = { x: i(t, "x", "neg") };
-  return o.runKernel(ee, s);
-}
-const be = /* @__PURE__ */ h({ neg_: De });
-/**
- * @license
- * Copyright 2020 Google LLC. All Rights Reserved.
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- * =============================================================================
- */
-function ye(t, e) {
-  const s = i(t, "x", "prelu"), n = i(e, "alpha", "prelu"), r = { x: s, alpha: n };
-  return o.runKernel(te, r);
-}
-const Se = /* @__PURE__ */ h({ prelu_: ye });
-/**
- * @license
- * Copyright 2020 Google LLC. All Rights Reserved.
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- * =============================================================================
- */
-function Ke(t) {
-  const s = { input: i(t, "input", "real") };
-  return o.runKernel(se, s);
-}
-const _e = /* @__PURE__ */ h({ real_: Ke });
-/**
- * @license
- * Copyright 2020 Google LLC. All Rights Reserved.
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- * =============================================================================
- */
-function Me(t) {
-  const s = { x: i(t, "x", "relu") };
-  return o.runKernel(ne, s);
-}
-const We = /* @__PURE__ */ h({ relu_: Me });
-/**
- * @license
- * Copyright 2020 Google LLC. All Rights Reserved.
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- * =============================================================================
- */
-function we(t) {
-  const s = { x: i(t, "x", "relu6") };
-  return o.runKernel(re, s);
-}
-const ze = /* @__PURE__ */ h({ relu6_: we });
-/**
- * @license
- * Copyright 2018 Google LLC. All Rights Reserved.
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- * =============================================================================
- */
-function Ee(t, e = 0) {
-  const n = { x: i(t, "x", "step") }, r = { alpha: e };
-  return o.runKernel(ue, n, r);
-}
-const Oe = /* @__PURE__ */ h({ step_: Ee });
-/**
- * @license
- * Copyright 2018 Google LLC. All Rights Reserved.
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- * =============================================================================
- */
-function Fe(t, e, s) {
-  const n = i(t, "x", "transpose");
-  if (e == null && (e = n.shape.map((l, p) => p).reverse()), L(n.rank === e.length, () => `Error in transpose: rank of input ${n.rank} must match length of perm ${e}.`), e.forEach((l) => {
-    L(l >= 0 && l < n.rank, () => `All entries in 'perm' must be between 0 and ${n.rank - 1} but got ${e}`);
-  }), n.rank <= 1)
-    return n.clone();
-  const r = { x: n }, c = { perm: e };
-  return n.dtype === "complex64" ? ae(() => {
-    let l = _e(n), p = $e(n);
-    return l = o.runKernel(A, { x: l }, c), p = o.runKernel(A, { x: p }, c), s && (p = be(p)), pe(l, p);
-  }) : o.runKernel(A, r, c);
-}
-const Re = /* @__PURE__ */ h({ transpose_: Fe });
-/**
- * @license
- * Copyright 2019 Google LLC. All Rights Reserved.
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- * =============================================================================
- */
-function Ae(t, e, s) {
-  if (s == null || s === "linear")
-    return t;
-  if (s === "relu")
-    return ie(t, Oe(e));
-  throw new Error(`Cannot compute gradient for fused activation ${s}.`);
-}
-function Le(t, e) {
-  let s = e;
-  const n = oe(t.shape, e.shape);
-  return n.length > 0 && (s = ce(s, n)), f(s, t.shape);
-}
-function Te(t, e, s, n) {
-  if (e === "linear")
-    return t;
-  if (e === "relu")
-    return We(t);
-  if (e === "elu")
-    return me(t);
-  if (e === "relu6")
-    return ze(t);
-  if (e === "prelu")
-    return Se(t, s);
-  if (e === "leakyrelu")
-    return ke(t, n);
-  if (e === "sigmoid")
-    return fe(t);
-  throw new Error(`Unknown fused activation ${e}.`);
-}
-const Be = (t, e) => !(t > 0) || e === "linear";
-/**
- * @license
- * Copyright 2019 Google LLC. All Rights Reserved.
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- * =============================================================================
- */
-function Ne({ a: t, b: e, transposeA: s = !1, transposeB: n = !1, bias: r, activation: c = "linear", preluActivationWeights: l, leakyreluAlpha: p = 0.2 }) {
-  if (Be(o.state.gradientDepth, c) === !1) {
-    let x = m(t, e, s, n);
-    return r != null && (x = le(x, r)), Te(x, c, l, p);
-  }
-  let u = i(t, "a", "fused matMul"), a = i(e, "b", "fused matMul");
-  [u, a] = q(u, a);
-  const D = s ? u.shape[u.rank - 2] : u.shape[u.rank - 1], b = n ? a.shape[a.rank - 1] : a.shape[a.rank - 2], W = s ? u.shape[u.rank - 1] : u.shape[u.rank - 2], w = n ? a.shape[a.rank - 2] : a.shape[a.rank - 1], T = u.shape.slice(0, -2), y = a.shape.slice(0, -2), B = C(T), N = C(y);
-  L(D === b, () => `Error in fused matMul: inner shapes (${D}) and (${b}) of Tensors with shapes ${u.shape} and ${a.shape} and transposeA=${s} and transposeB=${n} must match.`);
-  const O = P(u.shape.slice(0, -2), a.shape.slice(0, -2)).concat([W, w]), F = s ? f(u, [B, D, W]) : f(u, [B, W, D]), R = n ? f(a, [N, w, b]) : f(a, [N, b, w]);
-  let S;
-  r != null && (S = i(r, "bias", "fused matMul"), [S] = q(S, u), P(O, S.shape));
-  let v;
-  l != null && (v = i(l, "prelu weights", "fused matMul"));
-  const G = (x, M) => {
-    const [g, $, k, z] = M, d = Ae(f(x, k.shape), k, c);
-    let K, _;
-    if (!s && !n ? (K = m(d, $, !1, !0), _ = m(g, d, !0, !1)) : !s && n ? (K = m(d, $, !1, !1), _ = m(d, g, !0, !1)) : s && !n ? (K = m($, d, !1, !0), _ = m(g, d, !1, !1)) : (K = m($, d, !0, !0), _ = m(d, g, !0, !0)), r != null) {
-      const V = Le(z, d);
-      return [K, _, V];
-    } else
-      return [K, _];
-  }, I = {
-    a: F,
-    b: R,
-    bias: S,
-    preluActivationWeights: v
-  }, j = { transposeA: s, transposeB: n, activation: c, leakyreluAlpha: p };
-  return r == null ? U((M, g, $) => {
-    const k = (
-      // tslint:disable-next-line: no-unnecessary-type-assertion
-      o.runKernel(H, I, j)
-    );
-    return $([M, g, k]), { value: f(k, O), gradFunc: G };
-  })(F, R) : U((M, g, $, k) => {
-    const z = (
-      // tslint:disable-next-line: no-unnecessary-type-assertion
-      o.runKernel(H, I, j)
-    );
-    return k([M, g, z, $]), { value: f(z, O), gradFunc: G };
-  })(F, R, S);
-}
-const Q = /* @__PURE__ */ h({ fusedMatMul_: Ne });
-/**
- * @license
- * Copyright 2018 Google LLC
- *
- * Use of this source code is governed by an MIT-style
- * license that can be found in the LICENSE file or at
- * https://opensource.org/licenses/MIT.
- * =============================================================================
- */
-class E extends Error {
-  constructor(e) {
-    super(e), Object.setPrototypeOf(this, E.prototype);
-  }
-}
-/**
- * @license
- * Copyright 2018 Google LLC
- *
- * Use of this source code is governed by an MIT-style
- * license that can be found in the LICENSE file or at
- * https://opensource.org/licenses/MIT.
- * =============================================================================
- */
-function ve(t, e, s, n) {
-  if (t.rank < 2 || e.rank < 2)
-    throw new E(`dot requires both inputs to be rank >= 2 but got x shape = ${t.shape} and y shape = ${e.shape}`);
-  if (e.rank >= 3) {
-    const r = t.shape.slice(-1)[0], c = e.shape.slice(-2)[0];
-    if (r !== c)
-      throw new E(`If rank y >= 3, then the second last dim of y must equal the last dim of x but got x shape = ${t.shape} and  y shape = ${e.shape}`);
-  }
-  if (t.rank === 2 && e.rank === 2)
-    return Q({
-      a: t,
-      b: e,
-      transposeA: !1,
-      transposeB: !1,
-      bias: null,
-      activation: s
-    });
-  {
-    const r = t.shape.slice(), c = r.pop();
-    t = f(t, [-1, c]);
-    const l = e.shape.slice(), p = l.pop(), u = l.pop(), a = [...l, p], D = Array.from({ length: e.rank }, (T, y) => y === 0 ? e.rank - 2 : y <= e.rank - 2 ? y - 1 : y);
-    e = f(Re(e, D), [u, -1]);
-    const b = [...r, ...a];
-    return f(Q({
-      a: t,
-      b: e,
-      transposeA: !1,
-      transposeB: !1,
-      bias: null,
-      activation: s
-    }), b);
-  }
-}
-class Ue {
+import { r as t, d as s } from "../random_width-oeUIlUZj.js";
+import "../index-pWA4_lUh.js";
+import { v as r } from "../variable-Dl_ub3pk.js";
+import { g as d } from "../gather-BPGW8RsB.js";
+class b {
   vocabSize;
   embedDim;
-  tf;
   tiedWeights;
   initializer;
-  constructor(e, s, n) {
-    this.vocabSize = s.vocabSize, this.embedDim = s.embedDim, this.tf = e, this.initializer = this.tf.initializers.randomNormal({
+  constructor(i, e) {
+    this.vocabSize = i.vocabSize, this.embedDim = i.embedDim, this.initializer = t({
       mean: 0,
       stddev: 0.02
-    }), this.tiedWeights = this.tf.variable(
+    }), this.tiedWeights = r(
       this.initializer.apply([this.vocabSize, this.embedDim]),
       !0,
-      n || "tied_embedding"
+      e || "tied_embedding"
     );
   }
   get variables() {
     return [this.tiedWeights];
   }
-  embed(e) {
-    return this.tf.gather(this.tiedWeights, e, 0);
+  embed(i) {
+    return d(this.tiedWeights, i, 0);
   }
-  project(e) {
-    return ve(e, this.tiedWeights.transpose());
+  project(i) {
+    return s(i, this.tiedWeights.transpose());
   }
   getWeights() {
     return [this.tiedWeights];
   }
-  setWeights(e) {
-    this.tiedWeights.assign(e[0]);
+  setWeights(i) {
+    this.tiedWeights.assign(i[0]);
   }
   getConfig() {
     return {
@@ -445,5 +43,5 @@ class Ue {
   }
 }
 export {
-  Ue as default
+  b as default
 };

package/dist/layers/TransformerBlock.d.ts

@@ -1,29 +1,28 @@
-import { default as TF } from '@tensorflow/tfjs';
 import { GPTConfig } from '../config';
 import { KVCache } from './CausalSelfAttention';
 import { default as RoPECache } from './RoPECache';
 import { default as MemoryProfiler } from '../utilities/profile';
 import { default as BaseLayer } from './BaseLayer';
+import { Tensor, Variable } from '@tensorflow/tfjs-core';
 export default class Block extends BaseLayer {
     private ln1;
     private attn;
     private ln2;
     private mlp;
-    private tf;
     private index;
     private _trainable;
     skipped: boolean;
-    constructor(tf: typeof TF, index: number, config: GPTConfig, ropeCache?: RoPECache);
+    constructor(index: number, config: GPTConfig, ropeCache?: RoPECache);
     setProfiler(value: MemoryProfiler | undefined): void;
-    get variables(): TF.Variable[];
+    get variables(): Variable[];
     get trainable(): boolean;
     set trainable(value: boolean);
-    saveWeights(map: Map<string, TF.Tensor[]>): void;
-    loadWeights(weights: Map<string, TF.Tensor[]>): void;
+    saveWeights(map: Map<string, Tensor[]>): void;
+    loadWeights(weights: Map<string, Tensor[]>): void;
     private getMLPOutput;
-    call(x: TF.Tensor, training?: boolean, includeAttention?: boolean, cache?: KVCache): {
-        output: TF.Tensor;
-        attention?: TF.Tensor;
+    call(x: Tensor, training?: boolean, includeAttention?: boolean, cache?: KVCache): {
+        output: Tensor;
+        attention?: Tensor;
         cache?: KVCache;
     };
     dispose(): void;

package/dist/layers/TransformerBlock.js

@@ -1,18 +1,18 @@
-import a from "./CausalSelfAttention.js";
+import h from "./CausalSelfAttention.js";
 import o from "./MLP.js";
 import r from "./RMSNorm.js";
 import p from "./BaseLayer.js";
-class f extends p {
+import { t as d } from "../index-pWA4_lUh.js";
+class g extends p {
   ln1;
   attn;
   ln2;
   mlp;
-  tf;
   index;
   _trainable = !0;
   skipped = !1;
-  constructor(t, i, s, e) {
-    super(), this.tf = t, this.index = i, this.ln1 = new r(t, [s.nEmbed], 1e-8, `block_${this.index}_rms1`), this.attn = new a(this.tf, this.index, s, e), this.ln2 = new r(t, [s.nEmbed], 1e-8, `block_${this.index}_rms2`), this.mlp = new o(this.tf, this.index, s);
+  constructor(t, s, i) {
+    super(), this.index = t, this.ln1 = new r([s.nEmbed], 1e-8, `block_${this.index}_rms1`), this.attn = new h(this.index, s, i), this.ln2 = new r([s.nEmbed], 1e-8, `block_${this.index}_rms2`), this.mlp = new o(this.index, s);
   }
   setProfiler(t) {
     this._profiler = t, this.attn.setProfiler(t), this.mlp.setProfiler(t), this.ln1.setProfiler(t), this.ln2.setProfiler(t);
@@ -37,17 +37,17 @@ class f extends p {
   loadWeights(t) {
     this.attn.loadWeights(t), this.mlp.loadWeights(t), this.ln1.setWeights(t.get(`block_${this.index}_rms1`) || []), this.ln2.setWeights(t.get(`block_${this.index}_rms2`) || []);
   }
-  getMLPOutput(t, i) {
-    const s = this.ln2.apply(t), e = this.mlp.call(s, i);
+  getMLPOutput(t, s) {
+    const i = this.ln2.apply(t), e = this.mlp.call(i, s);
     return t.add(e);
   }
-  call(t, i = !1, s = !1, e) {
-    return this.tf.tidy(() => {
+  call(t, s = !1, i = !1, e) {
+    return d(() => {
       if (this.skipped)
         return { output: t };
-      const l = this.ln1.apply(t), n = this.attn.call(l, i, s, e), h = t.add(n.output);
+      const l = this.ln1.apply(t), n = this.attn.call(l, s, i, e), a = t.add(n.output);
       return {
-        output: this.getMLPOutput(h, i),
+        output: this.getMLPOutput(a, s),
         attention: n.attention,
         cache: n.presentKV
       };
@@ -58,5 +58,5 @@ class f extends p {
   }
 }
 export {
-  f as default
+  g as default
 };

package/dist/main.d.ts

@@ -1,10 +1,12 @@
 export { default as NanoGPT } from './NanoGPTModel';
 export { default as TeachableLLM } from './TeachableLLM';
 export { default as CharTokeniser } from './tokeniser/CharTokeniser';
+export { default as BPETokeniser } from './tokeniser/bpe';
 export { default as waitForModel } from './utilities/waitForModel';
 export { default as loadTextData } from './data/textLoader';
 export type { ITrainerOptions } from './Trainer';
 export type { IGenerateOptions } from './Generator';
 export type { TrainingLogEntry } from './NanoGPTModel';
+export type { ITokeniser } from './tokeniser/type';
 export type { GPTConfig } from './config';
 export { estimateParameterCount, estimateMemoryUsage, estimateTrainingMemoryUsage, estimateResources, validateConfig, } from './utilities/parameters';

package/dist/main.js

@@ -1,23 +1,37 @@
-import { default as s } from "./NanoGPTModel.js";
-import { default as p } from "./TeachableLLM.js";
-import { default as d } from "./tokeniser/CharTokeniser.js";
-import { default as x } from "./utilities/waitForModel.js";
-import { default as T } from "./data/textLoader.js";
-import { estimateMemoryUsage as M, estimateParameterCount as C, estimateResources as c, estimateTrainingMemoryUsage as h, validateConfig as y } from "./utilities/parameters.js";
-import "./ops/scatterSub.js";
-import "./ops/gatherSub.js";
-import "./ops/attentionMask.js";
-import "./ops/qkv.js";
-import "./ops/rope.js";
+import { default as P } from "./NanoGPTModel.js";
+import { default as h } from "./TeachableLLM.js";
+import { default as y } from "./tokeniser/CharTokeniser.js";
+import { default as U } from "./tokeniser/bpe.js";
+import { default as v } from "./utilities/waitForModel.js";
+import { default as B } from "./data/textLoader.js";
+import { estimateMemoryUsage as E, estimateParameterCount as F, estimateResources as G, estimateTrainingMemoryUsage as N, validateConfig as R } from "./utilities/parameters.js";
+import "./index-pWA4_lUh.js";
+import "./ops/cpu/scatterSub.js";
+import "./ops/webgl/scatterSub.js";
+import "./ops/cpu/gatherSub.js";
+import "./ops/webgl/gatherSub.js";
+import "./ops/cpu/attentionMask.js";
+import "./ops/webgl/attentionMask.js";
+import "./ops/grads/attentionMask.js";
+import "./ops/cpu/qkv.js";
+import "./ops/webgl/qkv.js";
+import "./ops/grads/qkv.js";
+import "@tensorflow/tfjs";
+import "./ops/cpu/rope.js";
+import "./ops/webgl/rope.js";
+import "./ops/grads/rope.js";
+import "./ops/cpu/appendCache.js";
+import "./ops/webgl/appendCache.js";
 export {
-  d as CharTokeniser,
-  s as NanoGPT,
-  p as TeachableLLM,
-  M as estimateMemoryUsage,
-  C as estimateParameterCount,
-  c as estimateResources,
-  h as estimateTrainingMemoryUsage,
-  T as loadTextData,
-  y as validateConfig,
-  x as waitForModel
+  U as BPETokeniser,
+  y as CharTokeniser,
+  P as NanoGPT,
+  h as TeachableLLM,
+  E as estimateMemoryUsage,
+  F as estimateParameterCount,
+  G as estimateResources,
+  N as estimateTrainingMemoryUsage,
+  B as loadTextData,
+  R as validateConfig,
+  v as waitForModel
 };

package/dist/{mat_mul-Bu7bhLms.js → mat_mul-D7_a4KJn.js}

@@ -1,4 +1,4 @@
-import { o as m, d as s, f as c, E as M, B as f } from "./index-YPKosni4.js";
+import { o as m, h as s, p as c, E as M, B as p } from "./index-pWA4_lUh.js";
 /**
  * @license
  * Copyright 2020 Google LLC. All Rights Reserved.
@@ -15,13 +15,13 @@ import { o as m, d as s, f as c, E as M, B as f } from "./index-YPKosni4.js";
  * limitations under the License.
  * =============================================================================
  */
-function p(e, o, n = !1, l = !1) {
+function f(e, o, n = !1, l = !1) {
   let a = s(e, "a", "matMul"), t = s(o, "b", "matMul");
   [a, t] = c(a, t);
   const r = { a, b: t }, u = { transposeA: n, transposeB: l };
-  return M.runKernel(f, r, u);
+  return M.runKernel(p, r, u);
 }
-const i = /* @__PURE__ */ m({ matMul_: p });
+const h = /* @__PURE__ */ m({ matMul_: f });
 export {
-  i as m
+  h as m
 };