Langtrain
Langtrain Docs
DocsAPI ReferenceSDK Reference
AppChat
GitHubDiscord

Node.js SDK

The official Node.js SDK for training, deploying, and managing AI models with Langtrain.

Node 18+
TypeScript
Streaming

Installation

Install the SDK with npm, yarn, or pnpm. Node.js 18+ is required.
python
1npm install langtrain-ai
2

3# With yarn
4yarn add langtrain-ai
5

6# With pnpm
7pnpm add langtrain-ai

Authentication

Configure your API key to authenticate with Langtrain.
python
1import Langtrain from 'langtrain-ai';
2

3// Option 1: Environment variable (recommended)
4// LANGTRAIN_API_KEY=your-api-key
5

6// Option 2: Direct initialization
7const client = new Langtrain({
8  apiKey: 'your-api-key',
9});
10

11// Option 3: From environment
12const client = new Langtrain({
13  apiKey: process.env.LANGTRAIN_API_KEY,
14});

LoRA Training

Fine-tune models using LoRA for efficient, memory-friendly training.
python
1import { LoRATrainer } from 'langtrain-ai';
2

3const trainer = new LoRATrainer({
4  model: 'meta-llama/Llama-3.3-8B',
5  outputDir: './my-model',
6});
7

8// Train on your data
9await trainer.train('training_data.jsonl');
10

11// Save and upload the trained model
12await trainer.save();
13await trainer.push('my-custom-model');

Dataset Management

Upload and manage training datasets programmatically.
python
1import { Dataset } from 'langtrain-ai';
2

3// Upload a dataset
4const dataset = await Dataset.upload({
5  filePath: 'data.jsonl',
6  name: 'customer-support-v1',
7});
8

9console.log('Dataset ID:', dataset.id);
10console.log('Rows:', dataset.rowCount);
11

12// List all datasets
13const datasets = await Dataset.list();
14datasets.forEach(ds => {
15  console.log(`- ${ds.name} (${ds.status})`);
16});

Training Jobs

Create and monitor fine-tuning jobs on Langtrain Cloud.
python
1import { TrainingJob } from 'langtrain-ai';
2

3// Create a training job
4const job = await TrainingJob.create({
5  modelId: 'llama-3.3-8b',
6  datasetId: dataset.id,
7  config: {
8    method: 'qlora',
9    epochs: 3,
10    learningRate: 2e-4,
11    batchSize: 4,
12  },
13});
14

15// Poll for completion
16while (['pending', 'running'].includes(job.status)) {
17  await job.refresh();
18  console.log(`Status: ${job.status}, Progress: ${job.progress}%`);
19  await new Promise(r => setTimeout(r, 30_000));
20}
21

22console.log('Training completed:', job.modelId);

Inference

Generate text with your trained models.
python
1import { Model } from 'langtrain-ai';
2

3const model = await Model.load('my-custom-model');
4

5// Generate text
6const response = await model.generate({
7  prompt: 'Explain machine learning',
8  maxTokens: 200,
9  temperature: 0.7,
10});
11console.log(response);
12

13// Chat interface
14const chatResponse = await model.chat([
15  { role: 'user', content: 'Hello!' }
16]);
17console.log(chatResponse.content);
18

19// Streaming
20for await (const chunk of model.stream('Tell me a story')) {
21  process.stdout.write(chunk);
22}

TypeScript Support

Full TypeScript types are included out of the box.
python
1import Langtrain, {
2  type TrainingConfig,
3  type GenerateOptions,
4  type ChatMessage,
5} from 'langtrain-ai';
6

7const config: TrainingConfig = {
8  method: 'qlora',
9  epochs: 3,
10  learningRate: 2e-4,
11};
12

13const messages: ChatMessage[] = [
14  { role: 'system', content: 'You are a helpful assistant.' },
15  { role: 'user', content: 'What is fine-tuning?' },
16];
17

18const client = new Langtrain();
19const response = await client.chat(messages, { model: 'my-model' });

Error Handling

Handle errors gracefully with typed exception classes.
python
1import {
2  LangtrainError,
3  AuthenticationError,
4  RateLimitError,
5  ValidationError,
6  NotFoundError,
7} from 'langtrain-ai';
8

9try {
10  const job = await TrainingJob.create({ ... });
11} catch (err) {
12  if (err instanceof AuthenticationError) {
13    console.error('Invalid API key');
14  } else if (err instanceof RateLimitError) {
15    console.error(`Rate limited — retry in ${err.retryAfter}s`);
16  } else if (err instanceof ValidationError) {
17    console.error('Invalid config:', err.message);
18  } else if (err instanceof NotFoundError) {
19    console.error('Model or dataset not found');
20  }
21}