Text Generation

Overview

This guide explains how to use the AI.ML Text API for chat-based completions, summarisation, and sentiment analysis.
It supports models like llama-3.1-8b-instruct and others via the /openai/chat/completions endpoint.

Base URL: https://api.ai.ml

Auth: Bearer token via header:
Authorization: Bearer <AIML_API_KEY>

Text API

Endpoint

POST /openai/chat/completions

Request Body (JSON)

Parameter	Type	Required	Default	Description
model	string	✓	—	Model name, e.g., `"llama-3.1-8b-instruct"`. Link to all models `<link to model list>`
messages	array	✓	—	Conversation history: list of `{ role, content }` objects. Roles: `system`, `user`, `assistant`.
temperature	number	✕	1	Controls randomness (0 = deterministic, >1 = more creative).
max_tokens	integer	✕	2048	Maximum tokens to generate.
top_p	number	✕	1	Nucleus sampling: considers tokens with cumulative probability ≤ `top_p`.
frequency_penalty	number	✕	0	Penalizes repetition of phrases. Range: -2 to 2.
presence_penalty	number	✕	0	Encourages new topics. Range: -2 to 2.
stop	string[]	✕	—	List of stop sequences to cut off output.
stream	boolean	✕	false	If true, streams tokens as they’re generated.

Example Usage

import OpenAI from 'openai';

const client = new OpenAI({
apiKey: process.env.AIML_API_KEY,
baseURL: "https://api.ai.ml/openai",
});

async function main() {
const completion = await client.chat.completions.create({
  model: "llama-3.1-8b-instruct",
  messages: [
    { role: "system", content: "You are a helpful assistant of user" },
    { role: "user", content: "Explain quantum computing in simple terms." }
  ],
  temperature: 0.7,
  max_tokens: 500,
});

console.log(completion.choices[0].message.content);
}

main();

Building Chat Application

To build a chat application:

Maintain conversation history in messages.
Alternate roles: system → user → assistant.
Send the entire history on each API call for context.

Example:


"messages": [
  { "role": "system", "content": "You are a customer support bot." },
  { "role": "user", "content": "I can’t log into my account." },
  { "role": "assistant", "content": "Have you tried resetting your password?" },
  { "role": "user", "content": "Yes, but it didn’t work." }
]

The assistant’s next reply will use this context.

Text Summarisation

Summarisation is done by giving summarisation instructions in the user message.

Example:


"messages": [
  { "role": "system", "content": "You are a summarization assistant. Summarize the news article in 3 bullet points" },
  { "role": "user", "content": "Summarize the following: <paste long article here>" }
]

Sentiment Analysis

Sentiment analysis is also prompt-driven.
Ask the model to classify sentiment as Positive, Negative, or Neutral.

Example:


"messages": [
  { "role": "system", "content": "You are a sentiment analysis assistant. Classify the reviews given as Positive, Negative, or Neutral" },
  { "role": "user", "content": "The product is amazing, I loved the fast delivery!" }
]

Response: Positive

Notes & Best Practices

Always specify a system role to set behavior (“You are a legal summarisation assistant”, etc.). For best practices in prompting refer to our detailed Guidelines for Text Generation reference
Use temperature=0 for deterministic outputs (good for analysis, classification).
Use temperature≈0.7 for creative writing and conversation.
Keep max_tokens aligned with your use case (short answers vs long summaries).
Sentiment/summary tasks work best when you clearly instruct the model.
Refer to our detailed Guideline for Text generation reference