Home/CSV / Excel to LLM Fine-Tuning JSONL Converter

CSV / Excel to LLM Fine-Tuning JSONL Converter

CSV / Excel to LLM Fine-Tuning JSONL Converter
Drag & Drop CSV or Excel file here or Click to upload
Supports .csv, .xlsx, .xls, .tsv — Processed locally in your browser

What is a JSONL File and Why is it Used for LLM Fine-Tuning?

A JSONL (JSON Lines) file is a text format where each individual line is a fully valid, self-contained JSON object, separated by a standard newline character (\n). JSONL is the industry-standard file format for uploading fine-tuning datasets to major Large Language Model (LLM) providers.

Unlike standard JSON files that enclose a massive array of objects inside square brackets (requiring parser libraries to load the entire file into RAM at once), JSONL files can be streamed line-by-line. This streamable parsing is crucial for machine learning training loops, which process millions of training examples in small batch increments to optimize GPU VRAM constraints.

LLM Fine-Tuning Dataset Formats Explained

Depending on the AI model architecture and provider platform, your dataset needs to follow one of two formatting rules:

1. Chat Completion Format (OpenAI ChatML & Llama 3)

This is the standard format for training conversational virtual agents. Each row in your spreadsheet represents a conversation dialogue containing a list of roles:

  • System Role: Sets the behavior, rules, tone, and constraints of the assistant.
  • User Role: Represents the prompt query or mock instructions inputted to the assistant.
  • Assistant Role: Represents the expected model response target the LLM is learning to imitate.

JSONL Schema Example:

{"messages": [{"role": "system", "content": "You are a helpful assistant."}, {"role": "user", "content": "Hello!"}, {"role": "assistant", "content": "Hi there!"}]}

2. Prompt-Completion Format (Legacy / Base Model Tuning)

Used for training base text completion models on unstructured raw data, code generation, or single-turn classifications:

{"prompt": "Generate a SQL query: Select users.", "completion": "SELECT * FROM users;"}

How to Upload Your Compiled JSONL File to Providers

After downloading your compiled `.jsonl` file, follow these steps to initiate your training job:

1. OpenAI API Dashboard

  1. Navigate to the OpenAI Developer Platform dashboard.
  2. Under the left sidebar menu, click on Storage or Files and upload your compiled `.jsonl` file.
  3. Go to the Fine-Tuning section, click Create, select your training model (e.g. gpt-4o-mini), choose the uploaded file as the training set, and click Start.

2. Google Cloud Vertex AI

  1. Upload your `.jsonl` file to a Google Cloud Storage (GCS) bucket.
  2. Navigate to Vertex AI Model Registry and click Pipelines / Tuning.
  3. Create a new Tuning job, link the input dataset path to your GCS file location, and initialize training.

Frequently Asked Questions (FAQ)

Q: Does this converter support large Excel sheets?
A: Yes! Since all processing is performed client-side using JavaScript, files are read directly from local RAM. The converter can process datasets containing tens of thousands of rows quickly without crashing the page.

Q: What if some cells in my spreadsheet are empty?
A: Empty spreadsheet cells will translate to empty strings in the generated JSON. If a system prompt column is selected, but specific rows contain blank cells, the static system prompt text will be injected as a fallback.

Q: Is my training dataset uploaded to RTSALL servers?
A: Absolutely not. The RTSALL JSONL Converter parses files using local HTML5 APIs. Zero bytes of your training data are sent over the internet, making it compliant with strict corporate IP regulations.

Loading