langtrain 0.1.3 → 0.1.6

package/package.json

@@ -1,10 +1,13 @@
 {
   "name": "langtrain",
-  "version": "0.1.3",
+  "version": "0.1.6",
   "description": "Unified JavaScript SDK for Langtrain Ecosystem",
   "main": "./dist/index.js",
   "module": "./dist/index.mjs",
   "types": "./dist/index.d.ts",
+  "bin": {
+    "langtrain": "./dist/cli.js"
+  },
   "scripts": {
     "build": "tsup",
     "dev": "tsup --watch",
@@ -26,9 +29,15 @@
   "author": "Langtrain AI",
   "license": "MIT",
   "devDependencies": {
-    "langvision": "file:../langvision/js",
+    "@types/node": "^25.2.3",
     "langtune": "file:../langtune/js",
+    "langvision": "file:../langvision/js",
     "tsup": "^8.0.2",
     "typescript": "^5.4.2"
+  },
+  "dependencies": {
+    "@clack/prompts": "^1.0.1",
+    "commander": "^14.0.3",
+    "kleur": "^4.1.5"
   }
 }

package/src/cli.ts

@@ -0,0 +1,213 @@
+#!/usr/bin/env node
+import { intro, outro, select, text, spinner, isCancel, cancel } from '@clack/prompts';
+import { bgCyan, black, red, green } from 'kleur/colors';
+import { Command } from 'commander';
+import { Langvision, Langtune } from './index';
+
+// Initialize clients
+const vision = new Langvision();
+const tune = new Langtune();
+
+async function main() {
+    const program = new Command();
+
+    program
+        .name('langtrain')
+        .description('Langtrain CLI for AI Model Fine-tuning and Generation')
+        .version('0.1.5');
+
+    program.action(async () => {
+        console.clear();
+        intro(`${bgCyan(black(' langtrain '))}`);
+
+        const operation = await select({
+            message: 'Select an operation:',
+            options: [
+                { value: 'tune-finetune', label: '🧠 Fine-tune Text Model (Langtune)' },
+                { value: 'tune-generate', label: '📝 Generate Text (Langtune)' },
+                { value: 'vision-finetune', label: '👁️ Fine-tune Vision Model (Langvision)' },
+                { value: 'vision-generate', label: '🖼️ Generate Vision Response (Langvision)' },
+                { value: 'exit', label: '🚪 Exit' }
+            ],
+        });
+
+        if (isCancel(operation) || operation === 'exit') {
+            outro('Goodbye!');
+            process.exit(0);
+        }
+
+        try {
+            if (operation === 'tune-finetune') {
+                await handleTuneFinetune();
+            } else if (operation === 'tune-generate') {
+                await handleTuneGenerate();
+            } else if (operation === 'vision-finetune') {
+                await handleVisionFinetune();
+            } else if (operation === 'vision-generate') {
+                await handleVisionGenerate();
+            }
+        } catch (error: any) {
+            outro(red(`Error: ${error.message}`));
+            process.exit(1);
+        }
+
+        outro(green('Operation completed successfully!'));
+    });
+
+    program.parse(process.argv);
+}
+
+// Handler for Langtune Fine-tuning
+async function handleTuneFinetune() {
+    const model = await text({
+        message: 'Enter base model (e.g., gpt-3.5-turbo):',
+        placeholder: 'gpt-3.5-turbo',
+        validate(value) {
+            if (!value || value.length === 0) return 'Value is required!';
+        },
+    });
+    if (isCancel(model)) cancel('Operation cancelled.');
+
+    const trainFile = await text({
+        message: 'Enter path to training file:',
+        placeholder: './data.jsonl',
+        validate(value) {
+            if (!value || value.length === 0) return 'Value is required!';
+        },
+    });
+    if (isCancel(trainFile)) cancel('Operation cancelled.');
+
+    const epochs = await text({
+        message: 'Num Epochs:',
+        placeholder: '3',
+        initialValue: '3'
+    });
+    if (isCancel(epochs)) cancel('Operation cancelled.');
+
+    const s = spinner();
+    s.start('Starting fine-tuning job...');
+
+    try {
+        // Check if FinetuneConfig types match what's needed.
+        // Casting to any to bypass strict type checking for this demo or ensure types are imported correctly.
+        // In a real scenario, we'd construct the full config object.
+        const config: any = {
+            model: model as string,
+            trainFile: trainFile as string,
+            preset: 'default', // simplified
+            epochs: parseInt(epochs as string),
+            batchSize: 1,
+            learningRate: 2e-5,
+            loraRank: 16,
+            outputDir: './output'
+        };
+
+        await tune.finetune(config);
+        s.stop(green('Fine-tuning job started!'));
+    } catch (e: any) {
+        s.stop(red('Failed to start job.'));
+        throw e;
+    }
+}
+
+// Handler for Langtune Generation
+async function handleTuneGenerate() {
+    const model = await text({
+        message: 'Enter model path:',
+        placeholder: './output/model',
+        initialValue: './output/model'
+    });
+    if (isCancel(model)) cancel('Operation cancelled');
+
+    const prompt = await text({
+        message: 'Enter prompt:',
+        placeholder: 'Hello world',
+    });
+    if (isCancel(prompt)) cancel('Operation cancelled');
+
+    const s = spinner();
+    s.start('Generating response...');
+
+    try {
+        const response = await tune.generate(model as string, { prompt: prompt as string });
+        s.stop('Generation complete');
+        intro('Response:');
+        console.log(response);
+    } catch (e: any) {
+        s.stop(red('Generation failed.'));
+        throw e;
+    }
+}
+
+// Handler for Langvision Fine-tuning
+async function handleVisionFinetune() {
+    const model = await text({
+        message: 'Enter base vision model:',
+        placeholder: 'llava-v1.5-7b',
+        initialValue: 'llava-v1.5-7b'
+    });
+    if (isCancel(model)) cancel('Operation cancelled');
+
+    const dataset = await text({
+        message: 'Enter dataset path:',
+        placeholder: './dataset',
+    });
+    if (isCancel(dataset)) cancel('Operation cancelled');
+
+    const epochs = await text({
+        message: 'Num Epochs:',
+        placeholder: '3',
+        initialValue: '3'
+    });
+
+    const s = spinner();
+    s.start('Starting vision fine-tuning...');
+
+    try {
+        const config: any = {
+            model: model as string,
+            dataset: dataset as string,
+            epochs: parseInt(epochs as string),
+            batchSize: 1,
+            learningRate: 2e-5,
+            loraRank: 16,
+            outputDir: './vision-output'
+        };
+        await vision.finetune(config);
+        s.stop(green('Vision fine-tuning started!'));
+    } catch (e: any) {
+        s.stop(red('Failed to start vision job.'));
+        throw e;
+    }
+}
+
+// Handler for Langvision Generation
+async function handleVisionGenerate() {
+    const model = await text({
+        message: 'Enter model path:',
+        placeholder: './vision-output/model',
+        initialValue: './vision-output/model'
+    });
+    if (isCancel(model)) cancel('Operation cancelled');
+
+    const prompt = await text({
+        message: 'Enter prompt/image path:', // Simplified for CLI
+        placeholder: 'Describe this image...',
+    });
+    if (isCancel(prompt)) cancel('Operation cancelled');
+
+    const s = spinner();
+    s.start('Generating vision response...');
+
+    try {
+        const response = await vision.generate(model as string, { prompt: prompt as string });
+        s.stop('Generation complete');
+        intro('Response:');
+        console.log(response);
+    } catch (e: any) {
+        s.stop(red('Generation failed.'));
+        throw e;
+    }
+}
+
+main().catch(console.error);

package/tsup.config.ts

@@ -1,7 +1,7 @@
 import { defineConfig } from 'tsup';
 
 export default defineConfig({
-    entry: ['src/index.ts'],
+    entry: ['src/index.ts', 'src/cli.ts'],
     format: ['cjs', 'esm'],
     dts: true,
     splitting: false,