feat(minor): save and load history to chat command (#71)

* feat: save and load history to `chat` command
* build: support patch bump of minor features
* fix: show correct cli command on Windows

Author: giladgd

.releaserc.json

@@ -7,6 +7,7 @@
         ["@semantic-release/commit-analyzer", {
             "preset": "angular",
             "releaseRules": [
+                {"type": "feat", "scope": "minor", "release": "patch"},
                 {"type": "docs", "scope": "README", "release": "patch"}
             ]
         }],

docs/guide/index.md

@@ -11,7 +11,7 @@ npm install --save node-llama-cpp
 
 > `node-llama-cpp` comes with pre-built binaries for macOS, Linux and Windows.
 >
-> If binaries are not available for your platform, it'll fallback to download the latest version of `llama.cpp` and build it from source with `cmake`.
+> If binaries are not available for your platform, it'll fallback to download a release of `llama.cpp` and build it from source with `cmake`.
 > To disable this behavior, set the environment variable `NODE_LLAMA_CPP_SKIP_DOWNLOAD` to `true`.
 
 ## CUDA and Metal support

src/cli/cli.ts

@@ -6,6 +6,7 @@ import yargs from "yargs";
 // eslint-disable-next-line node/file-extension-in-import
 import {hideBin} from "yargs/helpers";
 import fs from "fs-extra";
+import {cliBinName} from "../config.js";
 import {DownloadCommand} from "./commands/DownloadCommand.js";
 import {BuildCommand} from "./commands/BuildCommand.js";
 import {OnPostInstallCommand} from "./commands/OnPostInstallCommand.js";
@@ -19,6 +20,7 @@ const packageJson = fs.readJSONSync(path.join(__dirname, "..", "..", "package.js
 const yarg = yargs(hideBin(process.argv));
 
 yarg
+    .scriptName(cliBinName)
     .usage("Usage: $0 <command> [options]")
     .command(DownloadCommand)
     .command(BuildCommand)

src/cli/commands/ChatCommand.ts

@@ -5,14 +5,15 @@ import {CommandModule} from "yargs";
 import chalk from "chalk";
 import fs from "fs-extra";
 import withOra from "../../utils/withOra.js";
-import {defaultChatSystemPrompt} from "../../config.js";
+import {chatCommandHistoryFilePath, defaultChatSystemPrompt} from "../../config.js";
 import {LlamaChatPromptWrapper} from "../../chatWrappers/LlamaChatPromptWrapper.js";
 import {GeneralChatPromptWrapper} from "../../chatWrappers/GeneralChatPromptWrapper.js";
 import {ChatMLChatPromptWrapper} from "../../chatWrappers/ChatMLChatPromptWrapper.js";
 import {getChatWrapperByBos} from "../../chatWrappers/createChatWrapperByBos.js";
 import {ChatPromptWrapper} from "../../ChatPromptWrapper.js";
 import {FalconChatPromptWrapper} from "../../chatWrappers/FalconChatPromptWrapper.js";
 import {getIsInDocumentationMode} from "../../state.js";
+import {ReplHistory} from "../../utils/ReplHistory.js";
 import type {LlamaGrammar} from "../../llamaEvaluator/LlamaGrammar.js";
 
 const modelWrappers = ["auto", "general", "llamaChat", "chatML", "falconChat"] as const;
@@ -36,7 +37,8 @@ type ChatCommand = {
     penalizeRepeatingNewLine: boolean,
     repeatFrequencyPenalty?: number,
     repeatPresencePenalty?: number,
-    maxTokens: number
+    maxTokens: number,
+    noHistory: boolean
 };
 
 export const ChatCommand: CommandModule<object, ChatCommand> = {
@@ -176,19 +178,26 @@ export const ChatCommand: CommandModule<object, ChatCommand> = {
                 default: 0,
                 description: "Maximum number of tokens to generate in responses. Set to `0` to disable. Set to `-1` to set to the context size",
                 group: "Optional:"
+            })
+            .option("noHistory", {
+                alias: "nh",
+                type: "boolean",
+                default: false,
+                description: "Don't load or save chat history",
+                group: "Optional:"
             });
     },
     async handler({
         model, systemInfo, systemPrompt, prompt, wrapper, contextSize,
         grammar, jsonSchemaGrammarFile, threads, temperature, topK, topP,
         gpuLayers, repeatPenalty, lastTokensRepeatPenalty, penalizeRepeatingNewLine,
-        repeatFrequencyPenalty, repeatPresencePenalty, maxTokens
+        repeatFrequencyPenalty, repeatPresencePenalty, maxTokens, noHistory
     }) {
         try {
             await RunChat({
                 model, systemInfo, systemPrompt, prompt, wrapper, contextSize, grammar, jsonSchemaGrammarFile, threads, temperature, topK,
                 topP, gpuLayers, lastTokensRepeatPenalty, repeatPenalty, penalizeRepeatingNewLine, repeatFrequencyPenalty,
-                repeatPresencePenalty, maxTokens
+                repeatPresencePenalty, maxTokens, noHistory
             });
         } catch (err) {
             console.error(err);
@@ -201,7 +210,7 @@ export const ChatCommand: CommandModule<object, ChatCommand> = {
 async function RunChat({
     model: modelArg, systemInfo, systemPrompt, prompt, wrapper, contextSize, grammar: grammarArg,
     jsonSchemaGrammarFile: jsonSchemaGrammarFilePath, threads, temperature, topK, topP, gpuLayers, lastTokensRepeatPenalty, repeatPenalty,
-    penalizeRepeatingNewLine, repeatFrequencyPenalty, repeatPresencePenalty, maxTokens
+    penalizeRepeatingNewLine, repeatFrequencyPenalty, repeatPresencePenalty, maxTokens, noHistory
 }: ChatCommand) {
     const {LlamaChatSession} = await import("../../llamaEvaluator/LlamaChatSession.js");
     const {LlamaModel} = await import("../../llamaEvaluator/LlamaModel.js");
@@ -273,21 +282,32 @@ async function RunChat({
     // this is for ora to not interfere with readline
     await new Promise(resolve => setTimeout(resolve, 1));
 
-    const rl = readline.createInterface({
-        input: process.stdin,
-        output: process.stdout
-    });
+    const replHistory = await ReplHistory.load(chatCommandHistoryFilePath, !noHistory);
+
+    async function getPrompt() {
+        const rl = readline.createInterface({
+            input: process.stdin,
+            output: process.stdout,
+            history: replHistory.history.slice()
+        });
+
+        const res: string = await new Promise((accept) => rl.question(chalk.yellow("> "), accept));
+        rl.close();
+
+        return res;
+    }
 
     // eslint-disable-next-line no-constant-condition
     while (true) {
-        const input: string = initialPrompt != null
+        const input = initialPrompt != null
             ? initialPrompt
-            : await new Promise((accept) => rl.question(chalk.yellow("> "), accept));
+            : await getPrompt();
 
         if (initialPrompt != null) {
             console.log(chalk.green("> ") + initialPrompt);
             initialPrompt = null;
-        }
+        } else
+            await replHistory.add(input);
 
         if (input === ".exit")
             break;

src/config.ts

@@ -18,6 +18,7 @@ export const llamaBinsGrammarsDirectory = path.join(__dirname, "..", "llama", "g
 export const llamaCppDirectory = path.join(llamaDirectory, "llama.cpp");
 export const llamaCppGrammarsDirectory = path.join(llamaDirectory, "llama.cpp", "grammars");
 export const tempDownloadDirectory = path.join(os.tmpdir(), "node-llama-cpp", uuid.v4());
+export const chatCommandHistoryFilePath = path.join(os.homedir(), ".node-llama-cpp.chat_repl_history");
 export const usedBinFlagJsonPath = path.join(llamaDirectory, "usedBin.json");
 export const binariesGithubReleasePath = path.join(llamaDirectory, "binariesGithubRelease.json");
 export const currentReleaseGitBundlePath = path.join(llamaDirectory, "gitRelease.bundle");

src/utils/ReplHistory.ts

@@ -0,0 +1,85 @@
+import fs from "fs-extra";
+import {withLock} from "./withLock.js";
+
+type ReplyHistoryFile = {
+    history: string[]
+};
+
+const emptyHistory: ReplyHistoryFile = {
+    history: []
+};
+
+export class ReplHistory {
+    private readonly _filePath: string | null;
+    private _fileContent: ReplyHistoryFile;
+
+    private constructor(filePath: string | null, fileContent: ReplyHistoryFile) {
+        this._filePath = filePath;
+        this._fileContent = fileContent;
+    }
+
+    public async add(line: string) {
+        if (this._filePath == null) {
+            this._fileContent = this._addItemToHistory(line, this._fileContent);
+            return;
+        }
+
+        await withLock(this, "file", async () => {
+            try {
+                const json = parseReplJsonfile(await fs.readJSON(this._filePath!));
+                this._fileContent = this._addItemToHistory(line, json);
+
+                await fs.writeJSON(this._filePath!, this._fileContent, {
+                    spaces: 4
+                });
+            } catch (err) {}
+        });
+    }
+
+    public get history(): readonly string[] {
+        return this._fileContent.history;
+    }
+
+    private _addItemToHistory(item: string, fileContent: ReplyHistoryFile) {
+        const newHistory = fileContent.history.slice();
+        const currentItemIndex = newHistory.indexOf(item);
+
+        if (currentItemIndex !== -1)
+            newHistory.splice(currentItemIndex, 1);
+
+        newHistory.unshift(item);
+
+        return {
+            ...fileContent,
+            history: newHistory
+        };
+    }
+
+    public static async load(filePath: string, saveAndLoadHistory: boolean = true) {
+        if (!saveAndLoadHistory)
+            return new ReplHistory(null, {
+                history: []
+            });
+
+        try {
+            if (!(await fs.pathExists(filePath)))
+                await fs.writeJSON(filePath, emptyHistory, {
+                    spaces: 4
+                });
+
+            const json = parseReplJsonfile(await fs.readJSON(filePath));
+            return new ReplHistory(filePath, json);
+        } catch (err) {
+            return new ReplHistory(null, {
+                history: []
+            });
+        }
+    }
+}
+
+function parseReplJsonfile(file: unknown): ReplyHistoryFile {
+    if (typeof file !== "object" || file == null || !("history" in file) || !(file.history instanceof Array) || file.history.some((item) => typeof item !== "string"))
+        throw new Error("Invalid ReplyHistory file");
+
+    return file as ReplyHistoryFile;
+}