@genai-fi/nanogpt 0.6.2 → 0.6.3

package/dist/TeachableLLM.d.ts

@@ -22,12 +22,15 @@ export default class TeachableLLM {
     meta: TeachableLLMMeta;
     constructor(tokeniser?: ITokeniser, model?: NanoGPT);
     get vocab(): string[];
+    /** Model is fully loaded */
     get loaded(): boolean;
     get config(): GPTConfig;
     get model(): NanoGPT;
     get tokeniser(): ITokeniser;
     get status(): TeachableLLMStatus;
+    /** Model is both ready and not busy */
     get ready(): boolean;
+    get busy(): boolean;
     estimateTrainingMemoryUsage(batchSize: number): number;
     private setStatus;
     saveModel(options?: SaveOptions): Promise<Blob>;

package/dist/TeachableLLM.js

@@ -1,13 +1,13 @@
-import { defaultConfig as d } from "./config.js";
-import l from "./NanoGPTModel.js";
-import { saveModel as f } from "./utilities/save.js";
-import { loadModel as u } from "./loader/load.js";
-import p from "./Generator.js";
-import _ from "./Trainer.js";
-import { E as c } from "./index-Dwqa6Zy2.js";
+import { defaultConfig as _ } from "./config.js";
+import f from "./NanoGPTModel.js";
+import { saveModel as u } from "./utilities/save.js";
+import { loadModel as d } from "./loader/load.js";
+import l from "./Generator.js";
+import p from "./Trainer.js";
+import { E as g } from "./index-Dwqa6Zy2.js";
 import { dummyPassTrainAsync as m } from "./utilities/dummy.js";
-import g from "./tokeniser/CharTokeniser.js";
-import w from "./tokeniser/bpe.js";
+import c from "./tokeniser/CharTokeniser.js";
+import k from "./tokeniser/bpe.js";
 import "./papaparse.min-C8l2Kvo1.js";
 import "./index-Tf7vU29b.js";
 import "./jszip.min-CjP2V1VV.js";
@@ -43,9 +43,9 @@ import "./ops/cpu/normRMS.js";
 import "./ops/webgl/normRMS.js";
 import "./ops/grads/normRMS.js";
 import "./ops/webgl/log.js";
-import k from "./utilities/profile.js";
+import w from "./utilities/profile.js";
 class a {
-  ee = new c();
+  ee = new g();
   _config;
   _model;
   _tokeniser;
@@ -58,66 +58,71 @@ class a {
   get vocab() {
     return this._tokeniser?.getVocab() || [];
   }
+  /** Model is fully loaded */
   get loaded() {
     return !!this._model && !!this._tokeniser && !!this._config;
   }
   get config() {
     if (!this._config)
-      throw new Error("Model configuration is not initialized.");
+      throw new Error("configuration_not_initialized.");
     return this._config.gpt;
   }
   get model() {
     if (!this._model)
-      throw new Error("Model is not initialized.");
+      throw new Error("model_not_initialized.");
     return this._model;
   }
   get tokeniser() {
     if (!this._tokeniser)
-      throw new Error("Tokeniser is not initialized.");
+      throw new Error("tokeniser_not_initialized.");
     return this._tokeniser;
   }
   get status() {
     return this._status;
   }
+  /** Model is both ready and not busy */
   get ready() {
-    return this._status === "ready" && !!this._model && !!this._tokeniser && this.tokeniser.trained;
+    return this._status === "ready" && !!this._model && !!this._tokeniser;
+  }
+  get busy() {
+    return this._status === "busy" || this._status === "training";
   }
   estimateTrainingMemoryUsage(t) {
-    const e = this._memoryRequirements ?? { perBatch: 0, gradients: 0 }, r = e.perBatch * t, o = e.gradients;
-    return r * 0.66 + o * 4;
+    const e = this._memoryRequirements ?? { perBatch: 0, gradients: 0 }, i = e.perBatch * t, o = e.gradients;
+    return i * 0.66 + o * 4;
   }
   setStatus(t) {
     this._status !== t && (this._status = t, this.ee.emit("status", t));
   }
   saveModel(t) {
     if (!this._model || !this._tokeniser)
-      throw new Error("Model or tokeniser is not initialized.");
-    return f(this._model, this._tokeniser, {
+      throw new Error("model_or_tokeniser_not_initialized.");
+    return u(this._model, this._tokeniser, {
       ...t,
       name: t?.name || this.meta.name
     });
   }
   static loadModel(t) {
     const e = new a();
-    return u(t).then(({ model: r, tokeniser: o, name: s }) => {
-      e._model = r, e._tokeniser = o, e._config = r.config, s && (e.meta.name = s), e.setStatus("warmup"), m(r).then((i) => {
-        e._memoryRequirements = i, e.setStatus("ready"), e.ee.emit("loaded");
-      }).catch((i) => {
-        e.setStatus("error"), e.ee.emit("error", i);
+    return d(t).then(({ model: i, tokeniser: o, name: s }) => {
+      e._model = i, e._tokeniser = o, e._config = i.config, s && (e.meta.name = s), e.setStatus("warmup"), m(i).then((r) => {
+        e._memoryRequirements = r, e.setStatus("ready"), e.ee.emit("loaded");
+      }).catch((r) => {
+        e.setStatus("error"), e.ee.emit("error", r);
       });
-    }).catch((r) => {
-      e.setStatus("error"), e.ee.emit("error", r);
+    }).catch((i) => {
+      e.setStatus("error"), e.ee.emit("error", i);
     }), e;
   }
   static create(t, e = {}) {
-    const r = { ...d, ...e }, o = t === "char" ? new g(r.vocabSize) : new w(r.vocabSize), s = new l(r), i = new a(o, s);
-    return i.setStatus("warmup"), m(s).then((n) => {
-      i._memoryRequirements = n, i.tokeniser.trained ? (i.setStatus("ready"), i.ee.emit("loaded")) : (i.setStatus("awaitingTokens"), i.ee.emit("loaded"), i.tokeniser.once("trainStatus", (h) => {
-        h === "trained" && i.setStatus("ready");
+    const i = { ..._, ...e }, o = t === "char" ? new c(i.vocabSize) : new k(i.vocabSize), s = new f(i), r = new a(o, s);
+    return r.setStatus("warmup"), m(s).then((n) => {
+      r._memoryRequirements = n, r.tokeniser.trained ? (r.setStatus("ready"), r.ee.emit("loaded")) : (r.setStatus("awaitingTokens"), r.ee.emit("loaded"), r.tokeniser.once("trainStatus", (h) => {
+        h === "trained" && r.setStatus("ready");
       }));
     }).catch((n) => {
-      i.setStatus("error"), i.ee.emit("error", n);
-    }), i;
+      r.setStatus("error"), r.ee.emit("error", n);
+    }), r;
   }
   getProfiler() {
     return this._model?.getProfiler();
@@ -128,24 +133,22 @@ class a {
   set enableProfiler(t) {
     if (t) {
       if (!this._config)
-        throw new Error("Model is not initialized.");
-      this._config.layerConfig.profiler || (this._config.layerConfig.profiler = new k());
+        return;
+      this._config.layerConfig.profiler || (this._config.layerConfig.profiler = new w());
     } else
       this._config?.layerConfig.profiler && (this._config.layerConfig.profiler = void 0);
   }
   getNumParams() {
-    if (!this._model)
-      throw new Error("Model is not initialized.");
-    return this._model.getNumParams();
+    return this._model ? this._model.getNumParams() : 0;
   }
   trainer() {
     if (!this._model || !this._tokeniser)
-      throw new Error("Model or tokeniser is not initialized.");
-    const t = new _(this._model, this._tokeniser);
-    return t.on("start", () => this.setStatus("training")), t.on("stop", () => this.setStatus("ready")), t.on("log", async (e, r) => {
+      throw new Error("model_or_tokeniser_not_initialized.");
+    const t = new p(this._model, this._tokeniser);
+    return t.on("start", () => this.setStatus("training")), t.on("stop", () => this.setStatus("ready")), t.on("log", async (e, i) => {
       const o = this.ee.listeners("trainStep");
       for (const s of o)
-        await s(e, r);
+        await s(e, i);
     }), t;
   }
   train(t, e) {
@@ -160,7 +163,7 @@ class a {
   generator() {
     if (!this._model || !this._tokeniser)
       throw new Error("model_or_tokeniser_not_initialized.");
-    const t = new p(this._model, this._tokeniser);
+    const t = new l(this._model, this._tokeniser);
     return t.on("start", () => {
       this.status === "ready" && this.setStatus("busy");
     }), t.on("stop", () => {

package/dist/training/sparseCrossEntropy.js

@@ -1,35 +1,28 @@
-import { gatherSub as L } from "../ops/gatherSub.js";
-import { scatterSub as y } from "../ops/scatterSub.js";
-import { e as u, l as i, z as S, t as f, b as G } from "../index-BAzbokzv.js";
-import { s as v } from "../softmax-fsdtf6JC.js";
+import { gatherSub as x } from "../ops/gatherSub.js";
+import { scatterSub as L } from "../ops/scatterSub.js";
+import { l as C, t as u, z as E, b as G } from "../index-BAzbokzv.js";
+import { s as y } from "../softmax-fsdtf6JC.js";
 import { m as z } from "../max-DtlIuVeW.js";
-import { l as k } from "../log_sum_exp-YEo2h3gb.js";
-function F(a, s) {
-  return f(() => {
-    const e = a.shape[a.shape.length - 1], o = a.shape.slice(0, -1).reduce((d, c) => d * c, 1), p = a.shape.length > 2 ? a.reshape([o, e]) : a, n = s.shape.length > 1 ? s.reshape([o]).cast("int32") : s.cast("int32"), t = z(p, -1, !0), r = G(p, t), h = k(r, -1);
-    return L(h, n, r);
+import { l as v } from "../log_sum_exp-YEo2h3gb.js";
+function k(t, s) {
+  return u(() => {
+    const n = t.shape[t.shape.length - 1], c = t.shape.slice(0, -1).reduce((o, e) => o * e, 1), h = t.shape.length > 2 ? t.reshape([c, n]) : t, p = s.shape.length > 1 ? s.reshape([c]).cast("int32") : s.cast("int32"), r = z(h, -1, !0), a = G(h, r), m = v(a, -1);
+    return x(m, p, a);
   });
 }
-function j() {
-  return u().backendName === "tensorflow" ? i((s, e, m) => {
-    const o = s.shape.length > 2 ? s.reshape([-1, s.shape[s.shape.length - 1]]) : s, p = e.shape.length > 1 ? e.reshape([-1]).cast("int32") : e.cast("int32"), [n, t] = u().runKernel(
-      "NativeSparseSoftmaxCrossEntropy",
-      { logits: o, labels: p },
-      {}
-    );
-    return m([t.reshape(s.shape)]), { value: n, gradFunc: (r, h) => [h[0], S(e)] };
-  }) : i(
+function A() {
+  return C(
     // @ts-expect-error Invalid params
-    (s, e, m) => {
-      const o = s.shape[s.shape.length - 1], n = s.shape.slice(0, -1).reduce((d, c) => d * c, 1), t = s.reshape([n, o]), r = e.reshape([n]).cast("int32"), h = F(t, r);
-      return m([t, r]), t.dispose(), r.dispose(), { value: h, gradFunc: (d, c) => f(() => {
-        const g = c[0], b = c[1], x = v(g), C = y(x, b, d), E = S(e);
-        return [C.reshape(s.shape), E];
+    (s, n, d) => {
+      const c = s.shape[s.shape.length - 1], p = s.shape.slice(0, -1).reduce((o, e) => o * e, 1), r = s.reshape([p, c]), a = n.reshape([p]).cast("int32"), m = k(r, a);
+      return d([r, a]), r.dispose(), a.dispose(), { value: m, gradFunc: (o, e) => u(() => {
+        const S = e[0], f = e[1], b = y(S), l = L(b, f, o), g = E(n);
+        return [l.reshape(s.shape), g];
       }) };
     }
   );
 }
 export {
-  j as createSoftmaxCrossEntropyWithGrad,
-  F as sparseSoftmaxCrossEntropy
+  A as createSoftmaxCrossEntropyWithGrad,
+  k as sparseSoftmaxCrossEntropy
 };

package/package.json

@@ -1,6 +1,6 @@
 {
     "name": "@genai-fi/nanogpt",
-    "version": "0.6.2",
+    "version": "0.6.3",
     "type": "module",
     "main": "dist/main.js",
     "types": "dist/main.d.ts",

package/dist/ops/node/sparseCrossEntropy.d.ts

@@ -1 +0,0 @@
-export {};

package/dist/ops/node/sparseCrossEntropy.js

@@ -1,11 +0,0 @@
-import { r as o } from "../../index-BAzbokzv.js";
-function r(e) {
-  const { logits: t, labels: n } = e.inputs;
-  return e.backend.executeMultipleOutputs("SparseSoftmaxCrossEntropyWithLogits", [], [t, n], 2);
-}
-const s = {
-  kernelName: "NativeSparseSoftmaxCrossEntropy",
-  backendName: "tensorflow",
-  kernelFunc: r
-};
-o(s);