@genai-fi/nanogpt 1.3.0 → 1.4.0

package/dist/api/responses.js

@@ -29,13 +29,10 @@ var r = class {
 	}
 	async _runForRecord(e, t, n) {
 		let r = t.previous_response_id ? this._responses.get(t.previous_response_id) : void 0;
-		if (r) {
-			let e = r.generator.getConversation();
-			t.input && Array.isArray(t.input) ? t.input = [...e, ...t.input] : t.input && typeof t.input == "string" ? t.input = [...e, {
-				role: t.nonConversational ? "text" : "user",
-				content: t.input
-			}] : t.input = e;
-		}
+		r && (t.input && Array.isArray(t.input) || t.input && typeof t.input == "string" && (t.input = [{
+			role: t.nonConversational ? "text" : "user",
+			content: t.input
+		}]));
 		let i = r ? r.generator : this._responses.get(e).generator, a = n ? () => n({
 			output: i.getConversation(),
 			id: e,
@@ -63,11 +60,17 @@ var r = class {
 				done: !1
 			};
 			return c.then((n) => {
-				o.done = !0, t.done = !0, t.output = n, this._hookedResponses.delete(e), this._resumeWaiters.delete(e), a && i.off("tokens", a), this.ee.emit("done", e), this._processNextJob();
+				o.done = !0, t.done = !0, t.output = n, this._hookedResponses.delete(e), this._resumeWaiters.delete(e);
+				let r = n[0]?._id || e;
+				n.forEach((e) => {
+					e._id === void 0 && (e._id = r);
+				}), a && i.off("tokens", a), this.ee.emit("done", e), this._processNextJob();
 			}), t;
 		}
-		let l = await c;
-		return a && i.off("tokens", a), o.done = !0, this._hookedResponses.delete(e), this._resumeWaiters.delete(e), this.ee.emit("done", e), this._processNextJob(), {
+		let l = await c, u = l[0]?._id || e;
+		return l.forEach((e) => {
+			e._id === void 0 && (e._id = u);
+		}), a && i.off("tokens", a), o.done = !0, this._hookedResponses.delete(e), this._resumeWaiters.delete(e), this.ee.emit("done", e), this._processNextJob(), {
 			output: l,
 			id: e,
 			done: !0

package/dist/inference/Generator.js

@@ -187,7 +187,9 @@ var y = class extends e {
 		this.reset();
 	}
 	initialise(e, n) {
-		if (this.cache && n?.noCache && this.reset(), this.initialPrompt = e || null, this.lastToken === -1 ? this.outputConversation = (this.initialPrompt || []).slice() : e && e.length > this.outputConversation.length && (this.outputConversation = (this.initialPrompt || []).slice(), this.resetCache()), !this.cache && !n?.noCache && (this.model.config.modelType !== "GenAI_NanoGPT_v1" || this.model.config.useRope)) {
+		if (this.cache && n?.noCache && this.reset(), this.initialPrompt = e || null, this.lastToken === -1 && this.outputConversation.length === 0 ? this.outputConversation = this.initialPrompt || [] : e && e.length > 0 && (e.forEach((e) => {
+			this.outputConversation.push(e);
+		}), this.resetCache()), !this.cache && !n?.noCache && (this.model.config.modelType !== "GenAI_NanoGPT_v1" || this.model.config.useRope)) {
 			let e = Array(this.model.config.nLayer);
 			for (let t = 0; t < this.model.config.nLayer; t++) e[t] = {
 				k: void 0,

package/dist/inference/tokenisePrompt.js

@@ -3,7 +3,7 @@ import { I as e } from "../dist-Da20xy8E.js";
 async function t(t, n, r, i) {
 	if (r) {
 		let a = r.length > 0 && (r[r.length - 1].role === "text" || r[r.length - 1].role === "assistant"), o;
-		return i?.nonConversational ? o = a && i?.continuation ? [t.bosToken, ...t.encode(r[r.length - 1].content)] : t.encodeAsSequence(r, !0) : a && i?.continuation ? (o = t.encodeConversation(r, !0), o[o.length - 1] === t.getSpecialTokenIndex("<|assistant_start|>") && o[o.length - 2] === t.getSpecialTokenIndex("<|assistant_end|>") && (o.pop(), o.pop())) : o = t.encodeConversation(r, !0), o.length > n && (o = o.slice(-n)), e([o], [1, o.length], "int32");
+		return i?.nonConversational ? o = a && i?.continuation ? [t.bosToken, ...t.encode(r[r.length - 1].content)] : t.encodeAsSequence(r, !0) : a && i?.continuation ? (o = t.encodeConversation(r, !0), o[o.length - 1] === t.getSpecialTokenIndex("<|assistant_start|>") && o[o.length - 2] === t.getSpecialTokenIndex("<|assistant_end|>") && (o.pop(), o.pop())) : o = t.encodeConversation(r, !0), o.length > n && (o = o.slice(-n)), i?.debug && console.log("Tokenised prompt:", t.decode(o)), e([o], [1, o.length], "int32");
 	} else {
 		let n = i?.nonConversational ? void 0 : t.getSpecialTokenIndex("<|assistant_start|>"), r = n === void 0 ? [t.bosToken] : [t.bosToken, n];
 		return e([r], [1, r.length], "int32");

package/dist/inference/types.d.ts

@@ -17,6 +17,7 @@ export interface IGeneratorOutputInternal extends IGeneratorOutput {
     outputTensor: Tensor;
 }
 export interface GeneratorConversation extends Conversation {
+    _id?: string;
     _completed?: boolean;
     _timestamp?: number;
     _output?: IGeneratorOutput[];
@@ -45,6 +46,7 @@ export interface IGenerateOptions {
     continuation?: boolean; /** Continue the previous response without adding a new turn. */
     chunkSize?: number; /** Number of tokens to generate before returning a partial response. */
     background?: boolean; /** Run the generation in the background and return immediately. */
+    debug?: boolean;
     _onChunk?: (output: IGeneratorOutput) => void | Promise<void>;
 }
 export interface IGeneratorResponse {

package/package.json

@@ -1,6 +1,6 @@
 {
     "name": "@genai-fi/nanogpt",
-    "version": "1.3.0",
+    "version": "1.4.0",
     "type": "module",
     "main": "dist/main.js",
     "types": "dist/main.d.ts",