git-aicommit 5.2.1 → 7.0.1

package/README.md

@@ -9,7 +9,7 @@ Tired of writing commit messages? Let the computer do it for you!
 ## Installation
 
 ```bash
-npm install -g git-aicommit
+bun add -g git-aicommit
 ```
 
 ## Configuration
@@ -55,7 +55,7 @@ export default {
     ],
     diffFilter: 'ACMRTUXB',
     completionPromptParams: {
-        model: "gpt-3.5-turbo",
+        model: "gpt-4o-mini",
         temperature: 0.0,
         maxTokens: 1000,
     }
@@ -83,6 +83,4 @@ alias gai='git-aicommit'
 gai
 ```
 
-It that simple!
-    
-
+It's that simple!

package/autocommit.js

@@ -1,19 +1,10 @@
-#!/usr/bin/env node
+#!/usr/bin/env bun
 
 import { execSync, spawn } from "child_process";
 import rc from 'rc';
-import {
-    ChatPromptTemplate,
-    HumanMessagePromptTemplate,
-    PromptTemplate,
-    SystemMessagePromptTemplate
-} from "langchain/prompts";
 import defaultConfig from './config.js';
-import {RecursiveCharacterTextSplitter} from "langchain/text_splitter";
-import {loadSummarizationChain} from "langchain/chains";
-import {ChatOpenAI} from "langchain/chat_models/openai";
-import {OpenAI} from "langchain/llms/openai";
-import fs from "fs";
+import { OpenAI } from "openai";
+import {calculateMaxTokens, getModelContextSize} from "./count_tokens.js";
 
 const config = rc(
     'git-aicommit',
@@ -25,26 +16,26 @@ const config = rc(
 );
 
 try {
-  execSync(
-    'git rev-parse --is-inside-work-tree',
-    {encoding: 'utf8', stdio: 'ignore'}
-  );
+    execSync(
+        'git rev-parse --is-inside-work-tree',
+        {encoding: 'utf8', stdio: 'ignore'}
+    );
 } catch (e) {
-  console.error("This is not a git repository");
-  process.exit(1);
+    console.error("This is not a git repository");
+    process.exit(1);
 }
 
 if (!config.openAiKey && !config.azureOpenAiKey) {
-  console.error("Please set OPENAI_API_KEY or AZURE_OPENAI_API_KEY");
-  process.exit(1);
+    console.error("Please set OPENAI_API_KEY or AZURE_OPENAI_API_KEY");
+    process.exit(1);
 }
 
 // if any settings related to AZURE are set, if there are items that are not set, will error.
 if (config.azureOpenAiKey && !(
     config.azureOpenAiInstanceName && config.azureOpenAiDeploymentName && config.azureOpenAiVersion
 )){
-  console.error("Please set AZURE_OPENAI_API_KEY, AZURE_OPENAI_API_INSTANCE_NAME, AZURE_OPENAI_API_DEPLOYMENT_NAME, AZURE_OPENAI_API_VERSION when Azure OpenAI Service.");
-  process.exit(1);
+    console.error("Please set AZURE_OPENAI_API_KEY, AZURE_OPENAI_API_INSTANCE_NAME, AZURE_OPENAI_API_DEPLOYMENT_NAME, AZURE_OPENAI_API_VERSION when Azure OpenAI Service.");
+    process.exit(1);
 }
 
 const excludeFromDiff = config.excludeFromDiff || [];
@@ -53,8 +44,8 @@ const diffCommand = `git diff --staged \
     --no-ext-diff \
     --diff-filter=${diffFilter} \
     -- ${excludeFromDiff.map(
-        (pattern) => `':(exclude)${pattern}'`
-    ).join(' ')}
+    (pattern) => `':(exclude)${pattern}'`
+).join(' ')}
 `;
 
 let diff = execSync(diffCommand, {encoding: 'utf8'});
@@ -64,88 +55,47 @@ if (!diff) {
     process.exit(1);
 }
 
-const openai = new ChatOpenAI({
-    modelName: config.modelName,
-    openAIApiKey: config.openAiKey,
-    azureOpenAIApiKey: config.azureOpenAiKey,
-    azureOpenAIApiInstanceName: config.azureOpenAiInstanceName,
-    azureOpenAIApiDeploymentName: config.azureOpenAiDeploymentName,
-    azureOpenAIApiVersion: config.azureOpenAiVersion,
-    temperature: config.temperature,
-    maxTokens: config.maxTokens,
+const openai = new OpenAI({
+    apiKey: config.openAiKey,
+    baseURL: config.azureOpenAiKey ? `https://${config.azureOpenAiInstanceName}.openai.azure.com/openai/deployments/${config.azureOpenAiDeploymentName}` : undefined,
+    defaultHeaders: config.azureOpenAiKey ? { 'api-key': config.azureOpenAiKey } : undefined
 });
 
-const systemMessagePromptTemplate = SystemMessagePromptTemplate.fromTemplate(
-    config.systemMessagePromptTemplate
-);
-
-const humanPromptTemplate = HumanMessagePromptTemplate.fromTemplate(
-    config.humanPromptTemplate
-);
+async function getChatCompletion(messages) {
+    const response = await openai.chat.completions.create({
+        model: config.modelName || 'gpt-4o-mini',
+        messages: messages,
+        temperature: config.temperature,
+        max_tokens: config.maxTokens,
+    });
 
-const chatPrompt = ChatPromptTemplate.fromPromptMessages([
-    systemMessagePromptTemplate,
-    humanPromptTemplate,
-]);
-
-if (diff.length > 2000) {
-    console.log('Diff is too long. Splitting into multiple requests.')
-    const filenameRegex = /^a\/(.+?)\s+b\/(.+?)/;
-    const diffByFiles = diff
-        .split('diff ' + '--git ') // Wierd string concat in order to avoid splitting on this line when using autocommit in this repo :)
-        .filter((fileDiff) => fileDiff.length > 0)
-        .map((fileDiff) => {
-            const match = fileDiff.match(filenameRegex);
-            const filename = match ? match[1] : 'Unknown file';
-
-            const content = fileDiff
-                .replaceAll(filename, '')
-                .replaceAll('a/ b/\n', '')
-
-            return chatPrompt
-                .formatMessages({
-                    diff: content,
-                    language: config.language,
-                })
-                .then((prompt) => {
-                    return openai.call(prompt)
-                        .then((res) => {
-                            return {
-                                filename: filename,
-                                changes: res.text.trim(),
-                            }
-                        })
-                        .catch((e) => {
-                            console.error(`Error during OpenAI request: ${e.message}`);
-                            process.exit(1);
-                        });
-                });
-        });
-
-    // wait for all promises to resolve
-    const mergeChanges = await Promise.all(diffByFiles);
-
-    diff = mergeChanges
-        .map((fileDiff) => {
-            return `diff --git ${fileDiff.filename}\n${fileDiff.changes}`
-
-        })
-        .join('\n\n')
+    return response.choices[0].message.content.trim();
 }
 
-const prompt = await chatPrompt.formatMessages({
-    diff: diff,
-    language: config.language,
-})
+const systemMessage = { role: "system", content: config.systemMessagePromptTemplate };
+const userMessage = { role: "user", content: config.humanPromptTemplate.replace("{diff}", diff).replace("{language}", config.language) };
 
-const res = await openai.call(prompt)
-    .catch((e) => {
-        console.error(`Error during OpenAI request: ${e.message}`);
-        process.exit(1);
-    });
+const chatMessages = [systemMessage, userMessage];
+
+const tokenCount = await calculateMaxTokens({
+    prompt: diff,
+    modelName: config.modelName || 'gpt-4o-mini'
+});
+
+const contextSize = getModelContextSize(config.modelName || 'gpt-4o-mini');
+
+if (tokenCount > contextSize) {
+    console.log('Diff is too long. Please lower the amount of changes in the commit or switch to a model with bigger context size');
+
+    process.exit(1);
+}
 
-const commitMessage = res.text.trim();
+const messages = [
+    { role: "system", content: config.systemMessagePromptTemplate },
+    { role: "user", content: config.humanPromptTemplate.replace("{diff}", diff).replace("{language}", config.language) }
+];
 
+const commitMessage = await getChatCompletion(messages);
 
 if (!config.autocommit) {
     console.log(`Autocommit is disabled. Here is the message:\n ${commitMessage}`);

package/config.js

@@ -8,22 +8,21 @@ export default {
     openCommitTextEditor: false,
     language: 'english',
     systemMessagePromptTemplate: '' +
-        'You are expert AI, your job is to write clear and concise Git commit messages.' +
+        'You are expert software developer, your job is to write clear and concise Git commit messages. ' +
         'Your responsibility is to ensure that these messages accurately describe the changes made in each commit,' +
         'follow established guidelines. Provide a clear history of changes to the codebase.' +
         'Write 1-2 sentences. Output only the commit message without comments or other text.',
     humanPromptTemplate: '' +
         'Read the following git diff for a multiple files and ' +
         'write 1-2 sentences commit message in {language}' +
-        'without mentioning lines or files:\n' +
+        'without mentioning lines or files.' +
+        'If the reason behind the changed can be deducted from the changed, provide this reason:\n' +
         '{diff}',
     excludeFromDiff: [
-        '*.lock', '*.lockb'
+        '*.lock', '*.lockb', '*-lock.json', '*-lock.yaml'
     ],
     diffFilter: 'ACMRTUXB',
-    completionPromptParams: {
-        model: "gpt-3.5-turbo",
-        temperature: 0.0,
-        maxTokens: 1000,
-    }
+    modelName: "gpt-4o-mini",
+    temperature: 0.0,
+    maxTokens: 2000,
 }

package/count_tokens.js

@@ -0,0 +1,68 @@
+export const getModelNameForTiktoken = (modelName) => {
+    if (modelName.startsWith("gpt-3.5-turbo-16k")) {
+        return "gpt-3.5-turbo-16k";
+    }
+
+    if (modelName.startsWith("gpt-3.5-turbo-")) {
+        return "gpt-3.5-turbo";
+    }
+
+    if (modelName.startsWith("gpt-4-32k-")) {
+        return "gpt-4-32k";
+    }
+
+    if (modelName.startsWith("gpt-4-")) {
+        return "gpt-4";
+    }
+
+    if (modelName.startsWith("gpt-4o-")) {
+        return "gpt-4o";
+    }
+    return modelName;
+};
+
+
+export const getModelContextSize = (modelName) => {
+    switch (getModelNameForTiktoken(modelName)) {
+        case "gpt-3.5-turbo-16k":
+            return 16384;
+        case "gpt-3.5-turbo":
+            return 4096;
+        case "gpt-4-32k":
+            return 32768;
+        case "gpt-4":
+            return 8192;
+        case "gpt-4o":
+            return 128000;
+        default:
+            return 4096;
+    }
+};
+
+export const importTiktoken = async () => {
+    try {
+        const { encoding_for_model } = await import("@dqbd/tiktoken");
+        return { encoding_for_model };
+    } catch (error) {
+        console.log(error);
+        return { encoding_for_model: null };
+    }
+};
+
+export const calculateMaxTokens = async ({ prompt, modelName }) => {
+    const { encoding_for_model } = await importTiktoken();
+    // fallback to approximate calculation if tiktoken is not available
+    let numTokens = Math.ceil(prompt.length / 4);
+    try {
+        if (encoding_for_model) {
+            const encoding = encoding_for_model(getModelNameForTiktoken(modelName));
+            const tokenized = encoding.encode(prompt);
+            numTokens = tokenized.length;
+            encoding.free();
+        }
+    } catch (error) {
+        console.warn("Failed to calculate number of tokens with tiktoken, falling back to approximate count", error);
+    }
+    const maxTokens = getModelContextSize(modelName);
+    return maxTokens - numTokens;
+};

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "git-aicommit",
-  "version": "5.2.1",
+  "version": "7.0.1",
   "description": "Generates auto commit messages with OpenAI GPT3 model",
   "main": "autocommit.js",
   "repository": "https://github.com/shanginn/autocommit",
@@ -8,10 +8,9 @@
   "license": "MIT",
   "type": "module",
   "dependencies": {
-    "@dqbd/tiktoken": "^1.0.7",
-    "langchain": "^0.0.75",
-    "openai": "^3.2.1",
-    "rc": "^1.2.8"
+    "openai": "^4.52.7",
+    "rc": "^1.2.8",
+    "tiktoken": "^1.0.15"
   },
   "preferGlobal": true,
   "bin": {