@genai-fi/nanogpt 0.17.0 → 0.17.2

package/dist/TeachableLLM.d.ts

@@ -1,4 +1,4 @@
-import { GPTConfig } from './models/config';
+import { GPTConfig, LoRAConfig } from './models/config';
 import { Conversation, ITokeniser } from './tokeniser/type';
 import { SaveOptions } from './loader/save';
 import { LoadModelOptions } from './loader/load';
@@ -38,6 +38,13 @@ export default class TeachableLLM {
     /** Model is both ready and not busy */
     get ready(): boolean;
     get busy(): boolean;
+    createLoRA(name: string, loraConfig: LoRAConfig): void;
+    deleteLoRA(name: string): void;
+    renameLoRA(oldName: string, newName: string): void;
+    attachLoRA(name: string): void;
+    detachLoRA(): void;
+    hasLoRA(name?: string): boolean;
+    listLoRAs(): string[];
     estimateTrainingMemoryUsage(batchSize: number): number;
     private setStatus;
     saveModel(options?: SaveOptions): Promise<Blob>;
@@ -58,11 +65,11 @@ export default class TeachableLLM {
     on(event: 'phase', listener: (phase: ModelPhase) => void): void;
     on(event: 'error', listener: (error: Error) => void): void;
     on(event: 'trainStep', listener: (step: TrainingLogEntry) => void): void;
-    on(event: 'loaded', listener: () => void): void;
+    on(event: 'loaded' | 'changeLoRA', listener: () => void): void;
     off(event: 'status', listener: (status: TeachableLLMStatus) => void): void;
     off(event: 'phase', listener: (phase: ModelPhase) => void): void;
     off(event: 'error', listener: (error: Error) => void): void;
     off(event: 'trainStep', listener: (step: TrainingLogEntry) => void): void;
-    off(event: 'loaded', listener: () => void): void;
+    off(event: 'loaded' | 'changeLoRA', listener: () => void): void;
 }
 export {};

package/dist/TeachableLLM.js

@@ -1,7 +1,7 @@
 import { validateConfig as m } from "./models/config.js";
 import { saveModel as d } from "./loader/save.js";
-import { loadModel as p } from "./loader/load.js";
-import u from "./Generator.js";
+import { loadModel as u } from "./loader/load.js";
+import p from "./Generator.js";
 import h from "./Trainer.js";
 import { E as f } from "./index-DvYrXKkX.js";
 import { dummyPassTrainAsync as l } from "./utilities/dummy.js";
@@ -66,8 +66,8 @@ class a {
   _memoryRequirements;
   meta = {};
   _trainer = null;
-  constructor(t, r) {
-    this._config = r?.config, this._tokeniser = t, this._model = r;
+  constructor(t, i) {
+    this._config = i?.config, this._tokeniser = t, this._model = i;
   }
   get vocab() {
     return this._tokeniser?.getVocab() || [];
@@ -109,8 +109,43 @@ class a {
   get busy() {
     return this._status === "busy" || this._status === "training";
   }
+  createLoRA(t, i) {
+    if (!this._model)
+      throw new Error("model_not_initialized.");
+    this._model.createLoRA(t, i), this.ee.emit("changeLoRA");
+  }
+  deleteLoRA(t) {
+    if (!this._model)
+      throw new Error("model_not_initialized.");
+    this._model.deleteLoRA(t), this.ee.emit("changeLoRA");
+  }
+  renameLoRA(t, i) {
+    if (!this._model)
+      throw new Error("model_not_initialized.");
+    this._model.renameLoRA(t, i), this.ee.emit("changeLoRA");
+  }
+  attachLoRA(t) {
+    if (!this._model)
+      throw new Error("model_not_initialized.");
+    this._model.attachLoRA(t), this.ee.emit("changeLoRA");
+  }
+  detachLoRA() {
+    if (!this._model)
+      throw new Error("model_not_initialized.");
+    this._model.detachLoRA(), this.ee.emit("changeLoRA");
+  }
+  hasLoRA(t) {
+    if (!this._model)
+      throw new Error("model_not_initialized.");
+    return this._model.hasLoRA(t);
+  }
+  listLoRAs() {
+    if (!this._model)
+      throw new Error("model_not_initialized.");
+    return this._model.listLoRAs();
+  }
   estimateTrainingMemoryUsage(t) {
-    const r = this._memoryRequirements ?? { perBatch: 0, gradients: 0 }, e = r.perBatch * t, o = r.gradients;
+    const i = this._memoryRequirements ?? { perBatch: 0, gradients: 0 }, e = i.perBatch * t, o = i.gradients;
     return e * 0.66 + o * 4;
   }
   setStatus(t) {
@@ -124,10 +159,10 @@ class a {
       name: t?.name || this.meta.name
     });
   }
-  static loadModel(t, r) {
+  static loadModel(t, i) {
     const e = new a();
-    return p(t, r).then(({ model: o, tokeniser: n, metaData: i }) => {
-      m(o.config), e._model = o, e._tokeniser = n, e._config = o.config, i?.name && (e.meta.name = i.name), e.setStatus("warmup"), l(o).then((s) => {
+    return u(t, i).then(({ model: o, tokeniser: n, metaData: r }) => {
+      m(o.config), e._model = o, e._tokeniser = n, e._config = o.config, r?.name && (e.meta.name = r.name), e.setStatus("warmup"), l(o).then((s) => {
         e._memoryRequirements = s, e.setStatus("ready"), e.ee.emit("loaded"), e.ee.emit("phase", e.phase);
       }).catch((s) => {
         e.setStatus("error"), e.ee.emit("error", s), console.error("Error during warmup:", s);
@@ -136,16 +171,16 @@ class a {
       e.setStatus("error"), e.ee.emit("error", o), console.error("Error loading model:", o);
     }), e;
   }
-  static create(t, r) {
-    m(r);
-    const e = r, o = t === "char" ? new c(e.vocabSize) : new g(e.vocabSize), n = k(e), i = new a(o, n);
-    return i.setStatus("warmup"), l(n).then((s) => {
-      i._memoryRequirements = s, i.tokeniser.trained ? (i.setStatus("ready"), i.ee.emit("loaded"), i.ee.emit("phase", i.phase)) : (i.setStatus("awaitingTokens"), i.ee.emit("loaded"), i.ee.emit("phase", i.phase), i.tokeniser.once("trainStatus", (_) => {
-        _ === "trained" && i.setStatus("ready");
+  static create(t, i) {
+    m(i);
+    const e = i, o = t === "char" ? new c(e.vocabSize) : new g(e.vocabSize), n = k(e), r = new a(o, n);
+    return r.setStatus("warmup"), l(n).then((s) => {
+      r._memoryRequirements = s, r.tokeniser.trained ? (r.setStatus("ready"), r.ee.emit("loaded"), r.ee.emit("phase", r.phase)) : (r.setStatus("awaitingTokens"), r.ee.emit("loaded"), r.ee.emit("phase", r.phase), r.tokeniser.once("trainStatus", (_) => {
+        _ === "trained" && r.setStatus("ready");
       }));
     }).catch((s) => {
-      i.setStatus("error"), i.ee.emit("error", s), console.error("Error during warmup:", s);
-    }), i;
+      r.setStatus("error"), r.ee.emit("error", s), console.error("Error during warmup:", s);
+    }), r;
   }
   getProfiler() {
     return this._model?.getProfiler();
@@ -164,56 +199,56 @@ class a {
   getNumParams() {
     return this._model ? this._model.getNumParams() : 0;
   }
-  trainer(t, r) {
+  trainer(t, i) {
     if (!this._model || !this._tokeniser)
       throw new Error("model_or_tokeniser_not_initialized.");
     this._trainer && t && this._trainer.trainingType !== t && (this._trainer.dispose(), this._trainer = null);
-    const e = this._trainer === null ? new h(this._model, this._tokeniser, t, r) : new h(this._trainer, r);
+    const e = this._trainer === null ? new h(this._model, this._tokeniser, t, i) : new h(this._trainer, i);
     return e.on("start", () => {
       this.setStatus("training"), this.phase = t === "sft" ? "finetuned" : "pretrained";
     }), e.on("stop", () => this.setStatus("ready")), e.on("log", async (o) => {
       const n = this.ee.listeners("trainStep");
-      for (const i of n)
-        await i(o);
+      for (const r of n)
+        await r(o);
     }), this._trainer && this._trainer !== e && this._trainer.removeAllListeners(), this._trainer = e, e;
   }
-  async train(t, r, e) {
-    const o = this.trainer(e, r);
+  async train(t, i, e) {
+    const o = this.trainer(e, i);
     await o.prepare(t), await o.train();
   }
   async trainTokeniser(t) {
     if (!this._tokeniser)
       throw new Error("tokeniser_not_initialized.");
-    const r = await this._tokeniser.train(t);
-    return this._status === "awaitingTokens" && this.setStatus("ready"), r;
+    const i = await this._tokeniser.train(t);
+    return this._status === "awaitingTokens" && this.setStatus("ready"), i;
   }
   generator() {
     if (!this._model || !this._tokeniser)
       throw new Error("model_or_tokeniser_not_initialized.");
-    const t = new u(this._model, this._tokeniser);
+    const t = new p(this._model, this._tokeniser);
     return t.on("start", () => {
       this.status === "ready" && this.setStatus("busy");
     }), t.on("stop", () => {
       this.status === "busy" && this.setStatus("ready");
     }), t;
   }
-  generateText(t, r) {
-    return Array.isArray(t) ? this.generator().generate(t, r) : this.generator().generate([], r);
+  generateText(t, i) {
+    return Array.isArray(t) ? this.generator().generate(t, i) : this.generator().generate([], i);
   }
   dispose() {
     this._trainer && (this._trainer.dispose(), this._trainer = null), this._model?.dispose(), this.ee.removeAllListeners();
   }
   // eslint-disable-next-line @typescript-eslint/no-explicit-any
-  on(t, r) {
+  on(t, i) {
     if (t === "loaded" && this.loaded) {
-      setTimeout(() => r(), 0);
+      setTimeout(() => i(), 0);
       return;
     }
-    this.ee.on(t, r);
+    this.ee.on(t, i);
   }
   // eslint-disable-next-line @typescript-eslint/no-explicit-any
-  off(t, r) {
-    this.ee.off(t, r);
+  off(t, i) {
+    this.ee.off(t, i);
   }
 }
 export {

package/dist/models/model.js

@@ -58,15 +58,15 @@ class st extends i {
   loraLayer;
   loraMap = /* @__PURE__ */ new Map();
   constructor(t) {
-    super(t), t.loraConfig && (console.log(t.loraConfig), t.loraConfig.forEach((r, o) => {
+    super(t), t.loraConfig && t.loraConfig.forEach((r, o) => {
       this.createLoRA(o, r);
-    }));
+    });
   }
   createLoRA(t, r) {
     if (this.loraMap.has(t))
       return;
     const o = new e(t, this.weightStore, r.alpha, r.rank, r.variables);
-    this.loraMap.set(t, o), this.config.loraConfig = this.config.loraConfig || /* @__PURE__ */ new Map(), this.config.loraConfig.set(t, r), console.log(`Created LoRA ${t} with rank ${r.rank} and alpha ${r.alpha}`);
+    this.loraMap.set(t, o), this.config.loraConfig = this.config.loraConfig || /* @__PURE__ */ new Map(), this.config.loraConfig.set(t, r);
   }
   deleteLoRA(t) {
     const r = this.loraMap.get(t);

package/package.json

@@ -1,6 +1,6 @@
 {
     "name": "@genai-fi/nanogpt",
-    "version": "0.17.0",
+    "version": "0.17.2",
     "type": "module",
     "main": "dist/main.js",
     "types": "dist/main.d.ts",