README.md

metadata

license: apache-2.0
datasets:
  - PleIAs/common_corpus
language:
  - en
  - fr
  - es
  - de
  - it
  - la
  - nl
  - pl

Pleias-nano-1b-Preview is an early preview of a 1.21 billion parameters base model trained by Pleias with Tracto AI on Common Corpus.

Like all the base and specialized models from Pleias, Pleias-nano-1b-Preview has only been trained on open data out of copyright (public domain) or under a permissible license.

Description

Pleias-nano-1b-Preview is a transformer base model, entirely pretrained from scratch, using an architecture similar to Llama/GPT-Neox for easier deployment/inference.

It includes the following features, that would apply to any responsibly trained variant:

• Only trained on open data under a permissible license and in compliance with the European AI Act. By design, all Pleias model are unable to output copyrighted content.
• Extensive multilingual support for main European languages.
• A new tokenizer designed for enhanced document processing tasks and better multilingual support.
• Extremely low level of toxicity and problematic content.

Pleias-nano-1b-Preview has demonstrated unusual abilities for multilingual generation in its size range. Fully supported languages include English, French, Spanish, German, Italian, Dutch, Latin and Portuguese.

Given its size, Pleias-nano-1b-Preview can run on CPU without any compression loss. We provide a first GGUF variant as part of our release.

Recommended use

As a base model, Pleias-nano-1b-Preview is only able to run continuation prompts.

Text generation is currently able to support a range of creative writing tasks in multiple European languages. For more consistent results we recommend using a low or null temperature with a slight repetition penalty (1.2).

Pleias-nano-1b-Preview has been successfully adapted for continuous pretraining and full-fine-tuning on document processing tasks such as RAG, translation or OCR correction. Given the small size of the model we do not recommend fine-tuning methods based on LORA.

Example

Training

Pleias-nano-1b-Preview was fully pretrained on TractoAI on ISEG GPU cluster by Nebius AI on 192 h100s for 5 days. Pretraining code relied on the fork of Nanotron developed by TractoAI. We provide the complete settings as a yaml file as part of our release.

Training schedule includes 518,000 steps (batch size 1,024) on over three epochs (nearly 5 trillions tokens):

• A lightly filtered version of Common Corpus (1.6 trillion tokens)
• A filtered and enhanced version of Common Corpus (1,086,324,736,000 tokens).
• A repeat of the previous set.

Update

Pleias-nano-1b-Preview is currently released as an early preview.

The model will undergo several more round of post-training to enhance reasoning capacities and fine-tunability as well as in anticipation of a generalist instruct version.