node-llama-cpp 3.0.0-beta.9 → 3.0.1

package/dist/evaluator/LlamaChatSession/utils/LlamaChatSessionPromptCompletionEngine.js

@@ -0,0 +1,186 @@
+import { DisposeAggregator, DisposedError } from "lifecycle-utils";
+import { getConsoleLogPrefix } from "../../../utils/getConsoleLogPrefix.js";
+import { LruCache } from "../../../utils/LruCache.js";
+import { safeEventCallback } from "../../../utils/safeEventCallback.js";
+const defaultMaxPreloadTokens = 256;
+const defaultMaxCachedCompletions = 100;
+export class LlamaChatSessionPromptCompletionEngine {
+    /** @internal */ _chatSession;
+    /** @internal */ _maxPreloadTokens;
+    /** @internal */ _maxCachedCompletions;
+    /** @internal */ _onGeneration;
+    /** @internal */ _completionOptions;
+    /** @internal */ _completionCaches = new WeakMap();
+    /** @internal */ _disposeAggregator = new DisposeAggregator();
+    /** @internal */ _currentCompletionAbortController = new AbortController();
+    /** @internal */ _lastPrompt;
+    /** @internal */ _disposed = false;
+    constructor(chatSession, { maxPreloadTokens = defaultMaxPreloadTokens, onGeneration, maxCachedCompletions = defaultMaxCachedCompletions, ...options }) {
+        this._chatSession = chatSession;
+        this._maxPreloadTokens = Math.max(1, maxPreloadTokens);
+        this._maxCachedCompletions = Math.max(1, maxCachedCompletions);
+        this._onGeneration = safeEventCallback(onGeneration);
+        this._completionOptions = options;
+        this.dispose = this.dispose.bind(this);
+        this._disposeAggregator.add(this._chatSession.onDispose.createListener(this.dispose));
+        this._disposeAggregator.add(() => {
+            this._disposed = true;
+            this._currentCompletionAbortController.abort();
+        });
+    }
+    dispose() {
+        if (this._disposed)
+            return;
+        this._disposeAggregator.dispose();
+    }
+    /**
+     * Get completion for the prompt from the cache,
+     * and begin preloading this prompt into the context sequence and completing it.
+     *
+     * On completion progress, `onGeneration` (configured for this engine instance) will be called.
+     */
+    complete(prompt) {
+        if (this._disposed)
+            throw new DisposedError();
+        const completionCache = this._getCurrentCompletionCache();
+        const completion = completionCache.getCompletion(prompt);
+        if (this._lastPrompt == null || !(this._lastPrompt + (completion ?? "")).startsWith(prompt)) {
+            this._lastPrompt = prompt;
+            this._restartCompletion(completionCache);
+        }
+        this._lastPrompt = prompt;
+        return completion ?? "";
+    }
+    /** @internal */
+    _getCurrentCompletionCache() {
+        const completionCache = this._completionCaches.get(this._chatSession._chatHistoryStateRef);
+        if (completionCache != null)
+            return completionCache;
+        const newCompletionCache = new CompletionCache(this._maxCachedCompletions);
+        this._completionCaches.set(this._chatSession._chatHistoryStateRef, newCompletionCache);
+        return newCompletionCache;
+    }
+    /** @internal */
+    _restartCompletion(completionCache) {
+        if (this._disposed)
+            return;
+        this._currentCompletionAbortController.abort();
+        this._currentCompletionAbortController = new AbortController();
+        const prompt = this._lastPrompt;
+        if (prompt == null)
+            return;
+        const existingCompletion = completionCache.getCompletion(prompt);
+        const promptToComplete = prompt + (existingCompletion ?? "");
+        const currentPromptTokens = this._chatSession.model.tokenize(promptToComplete, false, "trimLeadingSpace").length;
+        const leftTokens = Math.max(0, this._maxPreloadTokens - currentPromptTokens);
+        if (leftTokens === 0)
+            return;
+        const currentAbortController = this._currentCompletionAbortController;
+        const currentAbortSignal = this._currentCompletionAbortController.signal;
+        let currentCompletion = "";
+        void this._chatSession.completePrompt(promptToComplete, {
+            ...this._completionOptions,
+            stopOnAbortSignal: false,
+            maxTokens: leftTokens,
+            signal: currentAbortSignal,
+            onTextChunk: (chunk) => {
+                currentCompletion += chunk;
+                const completion = (existingCompletion ?? "") + currentCompletion;
+                completionCache.putCompletion(prompt, completion);
+                if (this._getCurrentCompletionCache() !== completionCache) {
+                    currentAbortController.abort();
+                    return;
+                }
+                if (this._lastPrompt === prompt)
+                    this._onGeneration?.(prompt, completion);
+            }
+        })
+            .then(() => {
+            if (this._lastPrompt !== prompt && this._getCurrentCompletionCache() === completionCache)
+                return this._restartCompletion(completionCache);
+        })
+            .catch((err) => {
+            if (currentAbortSignal.aborted && err === currentAbortSignal.reason)
+                return;
+            console.error(getConsoleLogPrefix(false, false), err);
+        });
+    }
+    /** @internal */
+    static _create(chatSession, options = {}) {
+        return new LlamaChatSessionPromptCompletionEngine(chatSession, options);
+    }
+}
+class CompletionCache {
+    /** @internal */ _cache;
+    /** @internal */ _rootNode = [new Map()];
+    constructor(maxInputs) {
+        this._cache = new LruCache(maxInputs, {
+            onDelete: (key) => {
+                this._deleteInput(key);
+            }
+        });
+    }
+    get maxInputs() {
+        return this._cache.maxSize;
+    }
+    getCompletion(input) {
+        let node = this._rootNode;
+        for (let i = 0; i < input.length; i++) {
+            if (node == null)
+                return null;
+            const [next, completion] = node;
+            const char = input[i];
+            if (!next.has(char)) {
+                if (completion != null && completion.startsWith(input.slice(i))) {
+                    this._cache.get(input.slice(0, i));
+                    return completion.slice(input.length - i);
+                }
+            }
+            node = next.get(char);
+        }
+        if (node == null)
+            return null;
+        const [, possibleCompletion] = node;
+        if (possibleCompletion != null) {
+            this._cache.get(input);
+            return possibleCompletion;
+        }
+        return null;
+    }
+    putCompletion(input, completion) {
+        this._cache.set(input, null);
+        let node = this._rootNode;
+        for (let i = 0; i < input.length; i++) {
+            const [next] = node;
+            const char = input[i];
+            if (!next.has(char))
+                next.set(char, [new Map()]);
+            node = next.get(char);
+        }
+        const currentCompletion = node[1];
+        if (currentCompletion != null && currentCompletion.startsWith(completion))
+            return currentCompletion;
+        node[1] = completion;
+        return completion;
+    }
+    /** @internal */
+    _deleteInput(input) {
+        let lastNodeWithMultipleChildren = this._rootNode;
+        let lastNodeWithMultipleChildrenDeleteChar = input[0];
+        let node = this._rootNode;
+        for (let i = 0; i < input.length; i++) {
+            const [next] = node;
+            const char = input[i];
+            if (next.size > 1) {
+                lastNodeWithMultipleChildren = node;
+                lastNodeWithMultipleChildrenDeleteChar = char;
+            }
+            if (!next.has(char))
+                return;
+            node = next.get(char);
+        }
+        if (lastNodeWithMultipleChildrenDeleteChar !== "")
+            lastNodeWithMultipleChildren[0].delete(lastNodeWithMultipleChildrenDeleteChar);
+    }
+}
+//# sourceMappingURL=LlamaChatSessionPromptCompletionEngine.js.map

package/dist/evaluator/LlamaChatSession/utils/LlamaChatSessionPromptCompletionEngine.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"LlamaChatSessionPromptCompletionEngine.js","sourceRoot":"","sources":["../../../../src/evaluator/LlamaChatSession/utils/LlamaChatSessionPromptCompletionEngine.ts"],"names":[],"mappings":"AAAA,OAAO,EAAC,iBAAiB,EAAE,aAAa,EAAC,MAAM,iBAAiB,CAAC;AACjE,OAAO,EAAC,mBAAmB,EAAC,MAAM,uCAAuC,CAAC;AAC1E,OAAO,EAAC,QAAQ,EAAC,MAAM,4BAA4B,CAAC;AACpD,OAAO,EAAC,iBAAiB,EAAC,MAAM,qCAAqC,CAAC;AAkCtE,MAAM,uBAAuB,GAAG,GAAG,CAAC;AACpC,MAAM,2BAA2B,GAAG,GAAG,CAAC;AAExC,MAAM,OAAO,sCAAsC;IAC/C,gBAAgB,CAAkB,YAAY,CAAmB;IACjE,gBAAgB,CAAkB,iBAAiB,CAAS;IAC5D,gBAAgB,CAAkB,qBAAqB,CAAS;IAChE,gBAAgB,CAAkB,aAAa,CAA0D;IACzG,gBAAgB,CAAkB,kBAAkB,CAAiC;IACrF,gBAAgB,CAAkB,iBAAiB,GAAG,IAAI,OAAO,EAA2B,CAAC;IAC7F,gBAAgB,CAAkB,kBAAkB,GAAG,IAAI,iBAAiB,EAAE,CAAC;IAC/E,gBAAgB,CAAS,iCAAiC,GAAG,IAAI,eAAe,EAAE,CAAC;IACnF,gBAAgB,CAAS,WAAW,CAAU;IAC9C,gBAAgB,CAAS,SAAS,GAAG,KAAK,CAAC;IAE3C,YAAoB,WAA6B,EAAE,EAC/C,gBAAgB,GAAG,uBAAuB,EAC1C,YAAY,EACZ,oBAAoB,GAAG,2BAA2B,EAClD,GAAG,OAAO,EAC2B;QACrC,IAAI,CAAC,YAAY,GAAG,WAAW,CAAC;QAChC,IAAI,CAAC,iBAAiB,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,gBAAgB,CAAC,CAAC;QACvD,IAAI,CAAC,qBAAqB,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,oBAAoB,CAAC,CAAC;QAC/D,IAAI,CAAC,aAAa,GAAG,iBAAiB,CAAC,YAAY,CAAC,CAAC;QACrD,IAAI,CAAC,kBAAkB,GAAG,OAAO,CAAC;QAElC,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAEvC,IAAI,CAAC,kBAAkB,CAAC,GAAG,CACvB,IAAI,CAAC,YAAY,CAAC,SAAS,CAAC,cAAc,CAAC,IAAI,CAAC,OAAO,CAAC,CAC3D,CAAC;QACF,IAAI,CAAC,kBAAkB,CAAC,GAAG,CAAC,GAAG,EAAE;YAC7B,IAAI,CAAC,SAAS,GAAG,IAAI,CAAC;YACtB,IAAI,CAAC,iCAAiC,CAAC,KAAK,EAAE,CAAC;QACnD,CAAC,CAAC,CAAC;IACP,CAAC;IAEM,OAAO;QACV,IAAI,IAAI,CAAC,SAAS;YACd,OAAO;QAEX,IAAI,CAAC,kBAAkB,CAAC,OAAO,EAAE,CAAC;IACtC,CAAC;IAED;;;;;OAKG;IACI,QAAQ,CAAC,MAAc;QAC1B,IAAI,IAAI,CAAC,SAAS;YACd,MAAM,IAAI,aAAa,EAAE,CAAC;QAE9B,MAAM,eAAe,GAAG,IAAI,CAAC,0BAA0B,EAAE,CAAC;QAE1D,MAAM,UAAU,GAAG,eAAe,CAAC,aAAa,CAAC,MAAM,CAAC,CAAC;QAEzD,IAAI,IAAI,CAAC,WAAW,IAAI,IAAI,IAAI,CAAC,CAAC,IAAI,CAAC,WAAW,GAAG,CAAC,UAAU,IAAI,EAAE,CAAC,CAAC,CAAC,UAAU,CAAC,MAAM,CAAC,EAAE,CAAC;YAC1F,IAAI,CAAC,WAAW,GAAG,MAAM,CAAC;YAC1B,IAAI,CAAC,kBAAkB,CAAC,eAAe,CAAC,CAAC;QAC7C,CAAC;QAED,IAAI,CAAC,WAAW,GAAG,MAAM,CAAC;QAE1B,OAAO,UAAU,IAAI,EAAE,CAAC;IAC5B,CAAC;IAED,gBAAgB;IACR,0BAA0B;QAC9B,MAAM,eAAe,GAAG,IAAI,CAAC,iBAAiB,CAAC,GAAG,CAAC,IAAI,CAAC,YAAY,CAAC,oBAAoB,CAAC,CAAC;QAE3F,IAAI,eAAe,IAAI,IAAI;YACvB,OAAO,eAAe,CAAC;QAE3B,MAAM,kBAAkB,GAAG,IAAI,eAAe,CAAC,IAAI,CAAC,qBAAqB,CAAC,CAAC;QAC3E,IAAI,CAAC,iBAAiB,CAAC,GAAG,CAAC,IAAI,CAAC,YAAY,CAAC,oBAAoB,EAAE,kBAAkB,CAAC,CAAC;QACvF,OAAO,kBAAkB,CAAC;IAC9B,CAAC;IAED,gBAAgB;IACR,kBAAkB,CAAC,eAAgC;QACvD,IAAI,IAAI,CAAC,SAAS;YACd,OAAO;QAEX,IAAI,CAAC,iCAAiC,CAAC,KAAK,EAAE,CAAC;QAC/C,IAAI,CAAC,iCAAiC,GAAG,IAAI,eAAe,EAAE,CAAC;QAC/D,MAAM,MAAM,GAAG,IAAI,CAAC,WAAW,CAAC;QAEhC,IAAI,MAAM,IAAI,IAAI;YACd,OAAO;QAEX,MAAM,kBAAkB,GAAG,eAAe,CAAC,aAAa,CAAC,MAAM,CAAC,CAAC;QACjE,MAAM,gBAAgB,GAAG,MAAM,GAAG,CAAC,kBAAkB,IAAI,EAAE,CAAC,CAAC;QAE7D,MAAM,mBAAmB,GAAG,IAAI,CAAC,YAAY,CAAC,KAAK,CAAC,QAAQ,CAAC,gBAAgB,EAAE,KAAK,EAAE,kBAAkB,CAAC,CAAC,MAAM,CAAC;QACjH,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,iBAAiB,GAAG,mBAAmB,CAAC,CAAC;QAE7E,IAAI,UAAU,KAAK,CAAC;YAChB,OAAO;QAEX,MAAM,sBAAsB,GAAG,IAAI,CAAC,iCAAiC,CAAC;QACtE,MAAM,kBAAkB,GAAG,IAAI,CAAC,iCAAiC,CAAC,MAAM,CAAC;QACzE,IAAI,iBAAiB,GAAW,EAAE,CAAC;QACnC,KAAK,IAAI,CAAC,YAAY,CAAC,cAAc,CAAC,gBAAgB,EAAE;YACpD,GAAG,IAAI,CAAC,kBAAkB;YAC1B,iBAAiB,EAAE,KAAK;YACxB,SAAS,EAAE,UAAU;YACrB,MAAM,EAAE,kBAAkB;YAC1B,WAAW,EAAE,CAAC,KAAK,EAAE,EAAE;gBACnB,iBAAiB,IAAI,KAAK,CAAC;gBAC3B,MAAM,UAAU,GAAG,CAAC,kBAAkB,IAAI,EAAE,CAAC,GAAG,iBAAiB,CAAC;gBAClE,eAAe,CAAC,aAAa,CAAC,MAAM,EAAE,UAAU,CAAC,CAAC;gBAElD,IAAI,IAAI,CAAC,0BAA0B,EAAE,KAAK,eAAe,EAAE,CAAC;oBACxD,sBAAsB,CAAC,KAAK,EAAE,CAAC;oBAC/B,OAAO;gBACX,CAAC;gBAED,IAAI,IAAI,CAAC,WAAW,KAAK,MAAM;oBAC3B,IAAI,CAAC,aAAa,EAAE,CAAC,MAAM,EAAE,UAAU,CAAC,CAAC;YACjD,CAAC;SACJ,CAAC;aACG,IAAI,CAAC,GAAG,EAAE;YACP,IAAI,IAAI,CAAC,WAAW,KAAK,MAAM,IAAI,IAAI,CAAC,0BAA0B,EAAE,KAAK,eAAe;gBACpF,OAAO,IAAI,CAAC,kBAAkB,CAAC,eAAe,CAAC,CAAC;QACxD,CAAC,CAAC;aACD,KAAK,CAAC,CAAC,GAAG,EAAE,EAAE;YACX,IAAI,kBAAkB,CAAC,OAAO,IAAI,GAAG,KAAK,kBAAkB,CAAC,MAAM;gBAC/D,OAAO;YAEX,OAAO,CAAC,KAAK,CAAC,mBAAmB,CAAC,KAAK,EAAE,KAAK,CAAC,EAAE,GAAG,CAAC,CAAC;QAC1D,CAAC,CAAC,CAAC;IACX,CAAC;IAED,gBAAgB;IACT,MAAM,CAAC,OAAO,CAAC,WAA6B,EAAE,UAAkD,EAAE;QACrG,OAAO,IAAI,sCAAsC,CAAC,WAAW,EAAE,OAAO,CAAC,CAAC;IAC5E,CAAC;CACJ;AAED,MAAM,eAAe;IACjB,gBAAgB,CAAkB,MAAM,CAAyB;IACjE,gBAAgB,CAAkB,SAAS,GAAc,CAAC,IAAI,GAAG,EAAE,CAAC,CAAC;IAErE,YAAmB,SAAiB;QAChC,IAAI,CAAC,MAAM,GAAG,IAAI,QAAQ,CAAC,SAAS,EAAE;YAClC,QAAQ,EAAE,CAAC,GAAG,EAAE,EAAE;gBACd,IAAI,CAAC,YAAY,CAAC,GAAG,CAAC,CAAC;YAC3B,CAAC;SACJ,CAAC,CAAC;IACP,CAAC;IAED,IAAW,SAAS;QAChB,OAAO,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC;IAC/B,CAAC;IAEM,aAAa,CAAC,KAAa;QAC9B,IAAI,IAAI,GAA0B,IAAI,CAAC,SAAS,CAAC;QAEjD,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YACpC,IAAI,IAAI,IAAI,IAAI;gBACZ,OAAO,IAAI,CAAC;YAEhB,MAAM,CAAC,IAAI,EAAE,UAAU,CAAC,GAAc,IAAI,CAAC;YAC3C,MAAM,IAAI,GAAG,KAAK,CAAC,CAAC,CAAE,CAAC;YAEvB,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC;gBAClB,IAAI,UAAU,IAAI,IAAI,IAAI,UAAU,CAAC,UAAU,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;oBAC9D,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;oBACnC,OAAO,UAAU,CAAC,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;gBAC9C,CAAC;YACL,CAAC;YAED,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;QAC1B,CAAC;QAED,IAAI,IAAI,IAAI,IAAI;YACZ,OAAO,IAAI,CAAC;QAEhB,MAAM,CAAC,EAAE,kBAAkB,CAAC,GAAG,IAAI,CAAC;QACpC,IAAI,kBAAkB,IAAI,IAAI,EAAE,CAAC;YAC7B,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YACvB,OAAO,kBAAkB,CAAC;QAC9B,CAAC;QAED,OAAO,IAAI,CAAC;IAChB,CAAC;IAEM,aAAa,CAAC,KAAa,EAAE,UAAkB;QAClD,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,KAAK,EAAE,IAAI,CAAC,CAAC;QAE7B,IAAI,IAAI,GAAG,IAAI,CAAC,SAAS,CAAC;QAC1B,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YACpC,MAAM,CAAC,IAAI,CAAC,GAAG,IAAI,CAAC;YACpB,MAAM,IAAI,GAAG,KAAK,CAAC,CAAC,CAAE,CAAC;YAEvB,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC;gBACf,IAAI,CAAC,GAAG,CAAC,IAAI,EAAE,CAAC,IAAI,GAAG,EAAE,CAAC,CAAC,CAAC;YAEhC,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,CAAE,CAAC;QAC3B,CAAC;QAED,MAAM,iBAAiB,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC;QAClC,IAAI,iBAAiB,IAAI,IAAI,IAAI,iBAAiB,CAAC,UAAU,CAAC,UAAU,CAAC;YACrE,OAAO,iBAAiB,CAAC;QAE7B,IAAI,CAAC,CAAC,CAAC,GAAG,UAAU,CAAC;QACrB,OAAO,UAAU,CAAC;IACtB,CAAC;IAED,gBAAgB;IACR,YAAY,CAAC,KAAa;QAC9B,IAAI,4BAA4B,GAAc,IAAI,CAAC,SAAS,CAAC;QAC7D,IAAI,sCAAsC,GAAW,KAAK,CAAC,CAAC,CAAE,CAAC;QAE/D,IAAI,IAAI,GAAG,IAAI,CAAC,SAAS,CAAC;QAC1B,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YACpC,MAAM,CAAC,IAAI,CAAC,GAAG,IAAI,CAAC;YACpB,MAAM,IAAI,GAAG,KAAK,CAAC,CAAC,CAAE,CAAC;YAEvB,IAAI,IAAI,CAAC,IAAI,GAAG,CAAC,EAAE,CAAC;gBAChB,4BAA4B,GAAG,IAAI,CAAC;gBACpC,sCAAsC,GAAG,IAAI,CAAC;YAClD,CAAC;YAED,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC;gBACf,OAAO;YAEX,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,CAAE,CAAC;QAC3B,CAAC;QAED,IAAI,sCAAsC,KAAK,EAAE;YAC7C,4BAA4B,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,sCAAsC,CAAC,CAAC;IACvF,CAAC;CACJ"}

package/dist/evaluator/LlamaChatSession/utils/defineChatSessionFunction.d.ts

@@ -1,7 +1,15 @@
 import { GbnfJsonSchema, GbnfJsonSchemaToType } from "../../../utils/gbnfJson/types.js";
 import { ChatSessionModelFunction } from "../../../types.js";
+/**
+ * Define a function that can be used by the model in a chat session, and return it.
+ *
+ * This is a helper function to facilitate defining functions with full TypeScript type information.
+ *
+ * The handler function can return a Promise, and the return value will be awaited before being returned to the model.
+ * @param functionDefinition
+ */
 export declare function defineChatSessionFunction<const Params extends GbnfJsonSchema | undefined>({ description, params, handler }: {
     description?: string;
-    params?: Params;
-    handler: (params: GbnfJsonSchemaToType<Params>) => any;
+    params?: Params & GbnfJsonSchema;
+    handler: (params: GbnfJsonSchemaToType<Params>) => Promise<any> | any;
 }): ChatSessionModelFunction<Params>;

package/dist/evaluator/LlamaChatSession/utils/defineChatSessionFunction.js

@@ -1,3 +1,11 @@
+/**
+ * Define a function that can be used by the model in a chat session, and return it.
+ *
+ * This is a helper function to facilitate defining functions with full TypeScript type information.
+ *
+ * The handler function can return a Promise, and the return value will be awaited before being returned to the model.
+ * @param functionDefinition
+ */
 export function defineChatSessionFunction({ description, params, handler }) {
     return {
         description,

package/dist/evaluator/LlamaChatSession/utils/defineChatSessionFunction.js.map

@@ -1 +1 @@
-{"version":3,"file":"defineChatSessionFunction.js","sourceRoot":"","sources":["../../../../src/evaluator/LlamaChatSession/utils/defineChatSessionFunction.ts"],"names":[],"mappings":"AAGA,MAAM,UAAU,yBAAyB,CAAkD,EACvF,WAAW,EACX,MAAM,EACN,OAAO,EAKV;IACG,OAAO;QACH,WAAW;QACX,MAAM;QACN,OAAO;KACV,CAAC;AACN,CAAC"}
+{"version":3,"file":"defineChatSessionFunction.js","sourceRoot":"","sources":["../../../../src/evaluator/LlamaChatSession/utils/defineChatSessionFunction.ts"],"names":[],"mappings":"AAGA;;;;;;;GAOG;AACH,MAAM,UAAU,yBAAyB,CAAkD,EACvF,WAAW,EACX,MAAM,EACN,OAAO,EAKV;IACG,OAAO;QACH,WAAW;QACX,MAAM;QACN,OAAO;KACV,CAAC;AACN,CAAC"}

package/dist/evaluator/LlamaCompletion.d.ts

@@ -0,0 +1,168 @@
+import { EventRelay } from "lifecycle-utils";
+import { LLamaContextualRepeatPenalty, Token } from "../types.js";
+import { LlamaText } from "../utils/LlamaText.js";
+import { LlamaGrammar } from "./LlamaGrammar.js";
+import { EvaluationPriority } from "./LlamaContext/types.js";
+import { LlamaContextSequence } from "./LlamaContext/LlamaContext.js";
+import { TokenBias } from "./TokenBias.js";
+export type LlamaCompletionOptions = {
+    contextSequence: LlamaContextSequence;
+    /**
+     * Automatically dispose the sequence when the object is disposed.
+     *
+     * Defaults to `false`.
+     */
+    autoDisposeSequence?: boolean;
+};
+export type LlamaCompletionGenerationOptions = {
+    /**
+     * Called as the model generates a completion with the generated text chunk.
+     *
+     * Useful for streaming the generated completion as it's being generated.
+     */
+    onTextChunk?: (text: string) => void;
+    /**
+     * Called as the model generates a completion with the generated tokens.
+     *
+     * Preferably, you'd want to use `onTextChunk` instead of this.
+     */
+    onToken?: (tokens: Token[]) => void;
+    signal?: AbortSignal;
+    maxTokens?: number;
+    /**
+     * Temperature is a hyperparameter that controls the randomness of the generated text.
+     * It affects the probability distribution of the model's output tokens.
+     *
+     * A higher temperature (e.g., 1.5) makes the output more random and creative,
+     * while a lower temperature (e.g., 0.5) makes the output more focused, deterministic, and conservative.
+     *
+     * The suggested temperature is 0.8, which provides a balance between randomness and determinism.
+     *
+     * At the extreme, a temperature of 0 will always pick the most likely next token, leading to identical outputs in each run.
+     *
+     * Set to `0` to disable.
+     * Disabled by default (set to `0`).
+     */
+    temperature?: number;
+    /**
+     * From the next token candidates, discard the percentage of tokens with the lowest probability.
+     * For example, if set to `0.05`, 5% of the lowest probability tokens will be discarded.
+     * This is useful for generating more high-quality results when using a high temperature.
+     * Set to a value between `0` and `1` to enable.
+     *
+     * Only relevant when `temperature` is set to a value greater than `0`.
+     * Disabled by default.
+     */
+    minP?: number;
+    /**
+     * Limits the model to consider only the K most likely next tokens for sampling at each step of sequence generation.
+     * An integer number between `1` and the size of the vocabulary.
+     * Set to `0` to disable (which uses the full vocabulary).
+     *
+     * Only relevant when `temperature` is set to a value greater than 0.
+     */
+    topK?: number;
+    /**
+     * Dynamically selects the smallest set of tokens whose cumulative probability exceeds the threshold P,
+     * and samples the next token only from this set.
+     * A float number between `0` and `1`.
+     * Set to `1` to disable.
+     *
+     * Only relevant when `temperature` is set to a value greater than `0`.
+     */
+    topP?: number;
+    /**
+     * Used to control the randomness of the generated text.
+     *
+     * Change the seed to get different results.
+     *
+     * Only relevant when using `temperature`.
+     */
+    seed?: number;
+    /**
+     * Trim whitespace from the end of the generated text
+     * Disabled by default.
+     */
+    trimWhitespaceSuffix?: boolean;
+    repeatPenalty?: false | LLamaContextualRepeatPenalty;
+    /**
+     * Adjust the probability of tokens being generated.
+     * Can be used to bias the model to generate tokens that you want it to lean towards,
+     * or to avoid generating tokens that you want it to avoid.
+     */
+    tokenBias?: TokenBias | (() => TokenBias);
+    /**
+     * See the parameter `evaluationPriority` on the `LlamaContextSequence.evaluate()` function for more information.
+     */
+    evaluationPriority?: EvaluationPriority;
+    grammar?: LlamaGrammar;
+    /**
+     * Custom stop triggers to stop the completion when any of the provided triggers are found.
+     */
+    customStopTriggers?: readonly (LlamaText | string | readonly (string | Token)[])[];
+    /**
+     * The number of tokens to delete from the context window to make space for new ones.
+     * Defaults to 10% of the context size.
+     */
+    contextShiftSize?: number | ((sequence: LlamaContextSequence) => number | Promise<number>);
+    /**
+     * Context shift reconstructs the context with partial relevant data to continue generation when the context fills up.
+     * This flag disables this behavior.
+     * This flag will cause the generation to stop when the context fills up
+     * by setting an appropriate `maxTokens` value or lowering the given `maxTokens` value when needed.
+     * This flag will cause the generation to fail if there's no space for generating new tokens at all with the given inputs.
+     *
+     * Disabled by default. Not recommended unless you know what you're doing.
+     */
+    disableContextShift?: boolean;
+};
+export type LlamaInfillGenerationOptions = LlamaCompletionGenerationOptions & {
+    /**
+     * The minimum number of tokens to keep from the prefix input when making a context shift.
+     * Defaults to 10% of the context size.
+     */
+    minPrefixKeepTokens?: number | ((sequence: LlamaContextSequence) => number | Promise<number>);
+};
+export type LlamaCompletionResponse = {
+    response: string;
+    metadata: {
+        remainingGenerationAfterStop?: string | Token[];
+        stopReason: "eogToken" | "stopGenerationTrigger" | "maxTokens";
+    } | {
+        remainingGenerationAfterStop?: string | Token[];
+        stopReason: "customStopTrigger";
+        customStopTrigger: (string | Token)[];
+    };
+};
+export declare class LlamaCompletion {
+    readonly onDispose: EventRelay<void>;
+    constructor({ contextSequence, autoDisposeSequence }: LlamaCompletionOptions);
+    dispose({ disposeSequence }?: {
+        disposeSequence?: boolean;
+    }): void;
+    /** @hidden */
+    [Symbol.dispose](): void;
+    get disposed(): boolean;
+    get infillSupported(): boolean;
+    /**
+     * Generate a completion for an input.
+     */
+    generateCompletion(input: Token[] | string | LlamaText, options?: LlamaCompletionGenerationOptions): Promise<string>;
+    /**
+     * Same as `generateCompletion`, but returns additional metadata about the generation.
+     * See `generateCompletion` for more information.
+     */
+    generateCompletionWithMeta(input: Token[] | string | LlamaText, { onTextChunk, onToken, signal, maxTokens, temperature, minP, topK, topP, seed, trimWhitespaceSuffix, repeatPenalty, tokenBias, evaluationPriority, grammar, customStopTriggers, contextShiftSize, disableContextShift }?: LlamaCompletionGenerationOptions): Promise<LlamaCompletionResponse>;
+    /**
+     * Infill (also known as Fill-In-Middle), generates a completion for an input (`prefixInput`) that
+     * should connect to a given continuation (`suffixInput`).
+     * For example, for `prefixInput: "123"` and `suffixInput: "789"`, the model is expected to generate `456`
+     * to make the final text be `123456789`.
+     */
+    generateInfillCompletion(prefixInput: Token[] | string | LlamaText, suffixInput: Token[] | string | LlamaText, options?: LlamaInfillGenerationOptions): Promise<string>;
+    /**
+     * Same as `generateInfillCompletion`, but returns additional metadata about the generation.
+     * See `generateInfillCompletion` for more information.
+     */
+    generateInfillCompletionWithMeta(prefixInput: Token[] | string | LlamaText, suffixInput: Token[] | string | LlamaText, { onTextChunk, onToken, signal, maxTokens, temperature, minP, topK, topP, seed, trimWhitespaceSuffix, repeatPenalty, tokenBias, evaluationPriority, grammar, contextShiftSize, customStopTriggers, minPrefixKeepTokens, disableContextShift }?: LlamaInfillGenerationOptions): Promise<LlamaCompletionResponse>;
+}