feat: Add Input Token Limit Config Option (#281)

Author: mattsalt123

README.md

@@ -77,7 +77,8 @@ Create a `.env` file and add OpenCommit config variables there like this:
 
 ```env
 OCO_OPENAI_API_KEY=<your OpenAI API token>
-OCO_OPENAI_MAX_TOKENS=<max response tokens from OpenAI API>
+OCO_TOKENS_MAX_INPUT=<max model token limit (default: 4096)>
+OCO_TOKENS_MAX_OUTPUT=<max response tokens (default: 500)>
 OCO_OPENAI_BASE_PATH=<may be used to set proxy path to OpenAI api>
 OCO_DESCRIPTION=<postface a message with ~3 sentences description of the changes>
 OCO_EMOJI=<boolean, add GitMoji>
@@ -329,7 +330,8 @@ jobs:
           OCO_OPENAI_API_KEY: ${{ secrets.OCO_OPENAI_API_KEY }}
 
           # customization
-          OCO_OPENAI_MAX_TOKENS: 500
+          OCO_TOKENS_MAX_INPUT: 4096
+          OCO_TOKENS_MAX_OUTPUT: 500
           OCO_OPENAI_BASE_PATH: ''
           OCO_DESCRIPTION: false
           OCO_EMOJI: false

src/commands/config.ts

@@ -15,7 +15,8 @@ dotenv.config();
 
 export enum CONFIG_KEYS {
   OCO_OPENAI_API_KEY = 'OCO_OPENAI_API_KEY',
-  OCO_OPENAI_MAX_TOKENS = 'OCO_OPENAI_MAX_TOKENS',
+  OCO_TOKENS_MAX_INPUT = 'OCO_TOKENS_MAX_INPUT',
+  OCO_TOKENS_MAX_OUTPUT = 'OCO_TOKENS_MAX_OUTPUT',
   OCO_OPENAI_BASE_PATH = 'OCO_OPENAI_BASE_PATH',
   OCO_DESCRIPTION = 'OCO_DESCRIPTION',
   OCO_EMOJI = 'OCO_EMOJI',
@@ -26,13 +27,16 @@ export enum CONFIG_KEYS {
   OCO_AI_PROVIDER = 'OCO_AI_PROVIDER',
 }
 
-export const DEFAULT_MODEL_TOKEN_LIMIT = 4096;
-
 export enum CONFIG_MODES {
   get = 'get',
   set = 'set'
 }
 
+export enum DEFAULT_TOKEN_LIMITS {
+  DEFAULT_MAX_TOKENS_INPUT = 4096,
+  DEFAULT_MAX_TOKENS_OUTPUT = 500
+}
+
 const validateConfig = (
   key: string,
   condition: any,
@@ -75,18 +79,37 @@ export const configValidators = {
     return value;
   },
 
-  [CONFIG_KEYS.OCO_OPENAI_MAX_TOKENS](value: any) {
+  [CONFIG_KEYS.OCO_TOKENS_MAX_INPUT](value: any) {
+    // If the value is a string, convert it to a number.
+    if (typeof value === 'string') {
+      value = parseInt(value);
+      validateConfig(
+        CONFIG_KEYS.OCO_TOKENS_MAX_INPUT,
+        !isNaN(value),
+        'Must be a number'
+      );
+    }
+    validateConfig(
+      CONFIG_KEYS.OCO_TOKENS_MAX_INPUT,
+      value ? typeof value === 'number' : undefined,
+      'Must be a number'
+    );
+
+    return value;
+  },
+
+  [CONFIG_KEYS.OCO_TOKENS_MAX_OUTPUT](value: any) {
     // If the value is a string, convert it to a number.
     if (typeof value === 'string') {
       value = parseInt(value);
       validateConfig(
-        CONFIG_KEYS.OCO_OPENAI_MAX_TOKENS,
+        CONFIG_KEYS.OCO_TOKENS_MAX_OUTPUT,
         !isNaN(value),
         'Must be a number'
       );
     }
     validateConfig(
-      CONFIG_KEYS.OCO_OPENAI_MAX_TOKENS,
+      CONFIG_KEYS.OCO_TOKENS_MAX_OUTPUT,
       value ? typeof value === 'number' : undefined,
       'Must be a number'
     );
@@ -178,8 +201,11 @@ const configPath = pathJoin(homedir(), '.opencommit');
 export const getConfig = (): ConfigType | null => {
   const configFromEnv = {
     OCO_OPENAI_API_KEY: process.env.OCO_OPENAI_API_KEY,
-    OCO_OPENAI_MAX_TOKENS: process.env.OCO_OPENAI_MAX_TOKENS
-      ? Number(process.env.OCO_OPENAI_MAX_TOKENS)
+    OCO_TOKENS_MAX_INPUT: process.env.OCO_TOKENS_MAX_INPUT
+      ? Number(process.env.OCO_TOKENS_MAX_INPUT)
+      : undefined,
+    OCO_TOKENS_MAX_OUTPUT: process.env.OCO_TOKENS_MAX_OUTPUT
+      ? Number(process.env.OCO_TOKENS_MAX_OUTPUT)
       : undefined,
     OCO_OPENAI_BASE_PATH: process.env.OCO_OPENAI_BASE_PATH,
     OCO_DESCRIPTION: process.env.OCO_DESCRIPTION === 'true' ? true : false,

src/engine/openAi.ts

@@ -11,7 +11,7 @@ import { intro, outro } from '@clack/prompts';
 
 import {
   CONFIG_MODES,
-  DEFAULT_MODEL_TOKEN_LIMIT,
+  DEFAULT_TOKEN_LIMITS,
   getConfig
 } from '../commands/config';
 import { GenerateCommitMessageErrorEnum } from '../generateCommitMessageFromGitDiff';
@@ -20,7 +20,8 @@ import { AiEngine } from './Engine';
 
 const config = getConfig();
 
-let maxTokens = config?.OCO_OPENAI_MAX_TOKENS;
+const MAX_TOKENS_OUTPUT = config?.OCO_TOKENS_MAX_OUTPUT || DEFAULT_TOKEN_LIMITS.DEFAULT_MAX_TOKENS_OUTPUT;
+const MAX_TOKENS_INPUT = config?.OCO_TOKENS_MAX_INPUT || DEFAULT_TOKEN_LIMITS.DEFAULT_MAX_TOKENS_INPUT;
 let basePath = config?.OCO_OPENAI_BASE_PATH;
 let apiKey = config?.OCO_OPENAI_API_KEY
 
@@ -65,14 +66,14 @@ class OpenAi implements AiEngine {
       messages,
       temperature: 0,
       top_p: 0.1,
-      max_tokens: maxTokens || 500
+      max_tokens: MAX_TOKENS_OUTPUT
     };
     try {
       const REQUEST_TOKENS = messages
         .map((msg) => tokenCount(msg.content) + 4)
         .reduce((a, b) => a + b, 0);
 
-      if (REQUEST_TOKENS > DEFAULT_MODEL_TOKEN_LIMIT - maxTokens) {
+      if (REQUEST_TOKENS > MAX_TOKENS_INPUT - MAX_TOKENS_OUTPUT) {
         throw new Error(GenerateCommitMessageErrorEnum.tooMuchTokens);
       }
 

src/generateCommitMessageFromGitDiff.ts

@@ -3,13 +3,15 @@ import {
   ChatCompletionRequestMessageRoleEnum
 } from 'openai';
 
-import { DEFAULT_MODEL_TOKEN_LIMIT, getConfig } from './commands/config';
+import { DEFAULT_TOKEN_LIMITS, getConfig } from './commands/config';
 import { getMainCommitPrompt } from './prompts';
 import { mergeDiffs } from './utils/mergeDiffs';
 import { tokenCount } from './utils/tokenCount';
 import { getEngine } from './utils/engine';
 
 const config = getConfig();
+const MAX_TOKENS_INPUT = config?.OCO_TOKENS_MAX_INPUT || DEFAULT_TOKEN_LIMITS.DEFAULT_MAX_TOKENS_INPUT;
+const MAX_TOKENS_OUTPUT = config?.OCO_TOKENS_MAX_OUTPUT || DEFAULT_TOKEN_LIMITS.DEFAULT_MAX_TOKENS_OUTPUT;
 
 const generateCommitMessageChatCompletionPrompt = async (
   diff: string
@@ -29,7 +31,8 @@ const generateCommitMessageChatCompletionPrompt = async (
 export enum GenerateCommitMessageErrorEnum {
   tooMuchTokens = 'TOO_MUCH_TOKENS',
   internalError = 'INTERNAL_ERROR',
-  emptyMessage = 'EMPTY_MESSAGE'
+  emptyMessage = 'EMPTY_MESSAGE',
+  outputTokensTooHigh = `Token limit exceeded, OCO_TOKENS_MAX_OUTPUT must not be much higher than the default ${DEFAULT_TOKEN_LIMITS.DEFAULT_MAX_TOKENS_OUTPUT} tokens.`
 }
 
 const ADJUSTMENT_FACTOR = 20;
@@ -45,10 +48,10 @@ export const generateCommitMessageByDiff = async (
     ).reduce((a, b) => a + b, 0);
 
     const MAX_REQUEST_TOKENS =
-      DEFAULT_MODEL_TOKEN_LIMIT -
+      MAX_TOKENS_INPUT -
       ADJUSTMENT_FACTOR -
       INIT_MESSAGES_PROMPT_LENGTH -
-      config?.OCO_OPENAI_MAX_TOKENS;
+      MAX_TOKENS_OUTPUT;
 
     if (tokenCount(diff) >= MAX_REQUEST_TOKENS) {
       const commitMessagePromises = await getCommitMsgsPromisesFromFileDiffs(
@@ -124,6 +127,10 @@ function splitDiff(diff: string, maxChangeLength: number) {
   const splitDiffs = [];
   let currentDiff = '';
 
+  if (maxChangeLength <= 0) {
+    throw new Error(GenerateCommitMessageErrorEnum.outputTokensTooHigh);
+  }
+
   for (let line of lines) {
     // If a single line exceeds maxChangeLength, split it into multiple lines
     while (tokenCount(line) > maxChangeLength) {