alallo overview

What alallo is

alallo is a hosted service for fine-tuning open large language models (LLMs) on your own data. You choose a base model, a training method and your training data in the alallo dashboard. alallo then runs the training for you and gives you the fine-tuned model when the job finishes. You do not need to set up servers, install software or manage hardware to use alallo.

Who alallo is for

alallo is for teams and individuals who want a language model that follows their own style, knowledge or answer preferences, without running training infrastructure themselves. You do not need to be a machine learning expert to use alallo. The dashboard has sensible default settings for every training method, and the alallo help pages explain each setting in plain language.

What fine-tuning means

Fine-tuning means taking an existing, already-trained language model (the "base model") and training it further on your own examples. The fine-tuned model keeps the general abilities of the base model and learns the patterns in your data, for example your answer format, your terminology or which of two answers you prefer. alallo fine-tunes open models, such as the Qwen2.5 and Phi-3.5 families.

How an alallo job works, in short

An alallo training job goes through these steps:

1. You choose a base model, a training method (for example LoRA or DPO) and your training data (a public Hugging Face dataset or a file you upload).
2. You choose where your data may be processed (data residency: EU, US or Anywhere) and a goal (Lowest cost or Fastest finish).
3. You click "Estimate time and price". alallo reads your data, counts its tokens and shows a fixed price in alallo tokens and an estimated time range.
4. You start the job. alallo reserves the quoted tokens from your account's token balance.
5. You follow the job's status, progress, time left and training log in the dashboard.
6. When the job finishes, the fine-tuned model is ready and the quoted price is charged.

What you can try without an account

Visitors who are not signed in can browse the alallo dashboard: the available models, the training methods and their settings, and time and price estimates for public Hugging Face datasets. Uploading your own files and starting a training job require a free alallo account. Use "Sign up free" in the top bar of the dashboard to create one.

alallo help pages

• Getting started with alallo: create an account and run your first fine-tuning job.
• Models and training methods: the base models alallo offers, LoRA, QLoRA, full fine-tuning and DPO, and every training setting.
• Training data formats: uploading files, using Hugging Face datasets, the columns each method needs, size limits and validation errors.
• Tokens and pricing: the alallo token balance, top-ups, how a job's price is calculated, and the start fee per job.
• Cancelling jobs and refunds: what you pay when you cancel a job, when a cancel is free, and what happens when a job fails.
• Data residency and privacy: the EU, US and Anywhere options and how alallo handles your data.
• Accounts and users: the account owner (root) and member roles, inviting and removing users.
• Job outputs: what you get when a job finishes and how to use the fine-tuned model.
• Job status and time estimates: every job status and message, re-scheduling, and how time estimates work.
• Frequently asked questions: short answers to common questions about alallo.