Skip to main content

​
Formats

Nolano.AI supports specific data formats for different modalities to ensure seamless processing and training. Understanding these formats is crucial for preparing your data correctly.

​
Language & Code Data

​
JSONL Format

For language models and code generation, Nolano.AI uses JSONL (JSON Lines) format where each line contains a JSON object with a text field containing the training content. 
{"text": "def fibonacci(n):\n    if n <= 1:\n        return n\n    return fibonacci(n-1) + fibonacci(n-2)"}
{"text": "The quick brown fox jumps over the lazy dog. This is a sample text for language modeling."}
{"text": "import torch\nimport torch.nn as nn\n\nclass TransformerModel(nn.Module):\n    def __init__(self, vocab_size, d_model):\n        super().__init__()"}
{"text": "# Machine Learning Pipeline\n\nThis document explains how to build an ML pipeline using Python and scikit-learn."}
{"text": "SELECT customer_id, COUNT(*) as order_count FROM orders WHERE order_date >= '2023-01-01' GROUP BY customer_id"}

​
Best Practices

File Naming: Use .jsonl extension for clarity, though .json also works.
Escape Characters: Ensure proper escaping of newlines (\n), quotes (\"), and other special characters in JSON.

​
Time Series Data

​
AutoGluonTS Compatible Format

Time series data follows the AutoGluonTS format, where each line contains a JSON object with time series metadata and values.

​
Univariate Time Series

{"target": [1.2, 1.5, 1.8, 2.0, 1.9, 1.7, 2.1, 1.8], "start": "2023-01-01", "freq": "D"}
{"target": [45.2, 48.1, 52.3, 49.7, 51.2, 47.8, 50.1], "start": "2023-01-01", "freq": "H"}
{"target": [100, 105, 110, 108, 112, 115, 118, 120], "start": "2023-01-01 00:00:00", "freq": "15T"}
{"target": [23.5, 24.1, 23.8, 24.7, 25.2, 24.9, 25.5], "start": "2023-06-15", "freq": "D"}

​
Multivariate Time Series

Coming Soon: Multivariate time series support will be available in an upcoming release, enabling training on multiple correlated time series simultaneously.

​
Irregular Sample Intervals

Coming Soon: Support for irregular sample intervals will be added to handle time series data with non-uniform timestamps and varying frequencies.

​
Required Fields

​
target
array
required
Univariate: Array of numeric values representing the time seriesMultivariate: Array of arrays, where each inner array represents values at a time step across multiple dimensions
​
start
string
required
ISO format timestamp string indicating when the time series begins. Examples:
  • "2023-01-01" (date only)
  • "2023-01-01 00:00:00" (date and time)
  • "2023-01-01T00:00:00" (ISO 8601 format)
​
freq
string
required
Frequency string indicating the time interval between observations:
  • "D" - Daily
  • "H" - Hourly
  • "T" or "min" - Minutely
  • "S" - Secondly
  • "W" - Weekly
  • "M" - Monthly
  • "15T" - Every 15 minutes
  • "30S" - Every 30 seconds

​
Optional Fields

​
item_id
string
Unique identifier for the time series, useful when training on multiple series

​
Frequency Specification Guide

Common Frequencies

  • D - Daily
  • H - Hourly
  • T - Minutely
  • S - Secondly

Custom Intervals

  • 15T - Every 15 minutes
  • 30S - Every 30 seconds
  • 2H - Every 2 hours
  • 5D - Every 5 days

Business Frequencies

  • B - Business day
  • W - Weekly
  • M - Monthly
  • Q - Quarterly
Ready to prepare your data? Check out our Data Preparation Guide for step-by-step instructions on processing your data files.