@genai-fi/nanogpt 0.17.1 → 0.17.3

package/dist/Generator.d.ts

@@ -1,6 +1,10 @@
 import { Conversation, ITokeniser } from './tokeniser/type';
 import { default as EE } from 'eventemitter3';
 import { default as Model, ModelForwardAttributes } from './models/model';
+export interface GeneratorConversation extends Conversation {
+    _completed?: boolean;
+    _timestamp?: number;
+}
 export interface GenerateOptions {
     temperature?: number;
     topK?: number;

package/dist/Generator.js

@@ -11701,7 +11701,7 @@ class AS extends Ui {
     if (e) {
       const s = e.length > 0 && e[e.length - 1].role === "assistant";
       let o = [];
-      return n?.nonConversational ? s && n?.continuation ? o = t.encode(e[e.length - 1].content) : o = t.encodeAsSequence(e, !0) : o = t.encodeConversation(e, !0), o.length > this.model.config.blockSize && (o = o.slice(-this.model.config.blockSize)), Kt([o], [1, o.length], "int32");
+      return n?.nonConversational ? s && n?.continuation ? o = [t.bosToken, ...t.encode(e[e.length - 1].content)] : o = t.encodeAsSequence(e, !0) : o = t.encodeConversation(e, !0), o.length > this.model.config.blockSize && (o = o.slice(-this.model.config.blockSize)), Kt([o], [1, o.length], "int32");
     } else {
       const s = n?.nonConversational ? void 0 : t.getSpecialTokenIndex("<|assistant_start|>"), o = s !== void 0 ? [t.bosToken, s] : [t.bosToken];
       return Kt([o], [1, o.length], "int32");
@@ -11804,40 +11804,46 @@ class AS extends Ui {
   }
   /** Generate multiple tokens in a loop and produce text */
   async _generate(t) {
-    (this.lastToken < 0 || this.outputConversation.length === 0 || this.outputConversation[this.outputConversation.length - 1]._completed || this.outputConversation[this.outputConversation.length - 1].role !== "assistant") && (this.outputConversation.push({ role: "assistant", content: "", _timestamp: Date.now() }), this.resetCache(!t?.noCache));
-    let e = this.lastToken >= 0 && this.cache ? Kt([this.lastToken], [1, 1], "int32") : await this.tokenisePrompt(this.actualTokeniser, this.outputConversation.slice(0, -1), t);
-    const n = t?.maxLength ?? 1e3;
-    for (let s = 0; s < n && this.active; s++) {
+    let e = !1;
+    //this.lastToken < 0 ||
+    this.outputConversation.length === 0 || this.outputConversation[this.outputConversation.length - 1]._completed || this.outputConversation[this.outputConversation.length - 1].role !== "assistant" ? (this.outputConversation.push({ role: "assistant", content: "", _timestamp: Date.now() }), e = !0, this.resetCache(!t?.noCache)) : this.lastToken < 0 && this.resetCache(!t?.noCache);
+    let n = this.lastToken >= 0 && this.cache ? Kt([this.lastToken], [1, 1], "int32") : await this.tokenisePrompt(
+      this.actualTokeniser,
+      e ? this.outputConversation.slice(0, -1) : this.outputConversation,
+      t
+    );
+    const s = t?.maxLength ?? 1e3;
+    for (let o = 0; o < s && this.active; o++) {
       const {
-        output: o,
-        probabilities: r,
-        attention: c,
-        loss: i,
-        multinomialRand: l
-      } = await this._generateToken(e, this.cache ? this.cache : void 0, {
+        output: r,
+        probabilities: c,
+        attention: i,
+        loss: l,
+        multinomialRand: u
+      } = await this._generateToken(n, this.cache ? this.cache : void 0, {
         ...t,
         usePadding: !this.cache
       });
-      if (this.lastMultinomialRand = l, i !== void 0 && (this.lastLoss = i), this.cache)
-        e.dispose(), e = o;
+      if (this.lastMultinomialRand = u, l !== void 0 && (this.lastLoss = l), this.cache)
+        n.dispose(), n = r;
       else {
-        const p = e;
-        e = Gd([e, o], 1), p.dispose();
+        const d = n;
+        n = Gd([n, r], 1), d.dispose();
       }
-      const u = await this.processResponse(
+      const p = await this.processResponse(
         this.actualTokeniser,
-        o,
+        r,
         t?.allowSpecial ?? !1,
-        c,
-        r
+        i,
+        c
       );
-      if (this.cache || o.dispose(), u === null) {
+      if (this.cache || r.dispose(), p === null) {
         this.outputConversation[this.outputConversation.length - 1]._completed = !0;
         break;
       }
-      this.outputConversation[this.outputConversation.length - 1].content += u;
+      o === s - 1 && (this.outputConversation[this.outputConversation.length - 1]._completed = !0), this.outputConversation[this.outputConversation.length - 1].content += p;
     }
-    return e.dispose(), this.outputConversation;
+    return n.dispose(), this.outputConversation;
   }
   resetCache(t) {
     this.cache && (this.cache.forEach((e) => {

package/dist/TeachableLLM.d.ts

@@ -65,11 +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/package.json

@@ -1,6 +1,6 @@
 {
     "name": "@genai-fi/nanogpt",
-    "version": "0.17.1",
+    "version": "0.17.3",
     "type": "module",
     "main": "dist/main.js",
     "types": "dist/main.d.ts",