Pleias pretrained its Nano 1.2B preview model on TractoAI using Nebius AI’s ISEG GPU cluster. Its model card reports 192 H100 GPUs over five days. The training code used TractoAI’s fork of Nanotron. The training schedule comprised 518,000 steps with a batch size of 1,024, covering nearly five trillion tokens over three epochs. Pleias estimated training emissions at four tonnes CO2eq. Pleias also continuously pretrained its Nano RAG variant on TractoAI using the ISEG cluster, with a retrieval dataset containing about 45 billion tokens.
Pleias converted tokenized training data into TractoAI tables, built a Nanotron dataset adapter and adapted file operations, including checkpoint saving. It used tractorun to deploy and coordinate distributed training automatically with fault tolerance. Pleias also generated synthetic RAG and instruction training data with larger models through a Map-Reduce based TractoAI method.
For Common Corpus, Pleias credited Nebius AI and Tracto AI among the organizations supporting storage and processing. Its release article describes a multilingual training dataset containing more than two trillion permissibly licensed tokens. For Common Corpus 2, Pleias processed 11 million scientific articles from OpenAlex, representing about 175 billion tokens, using PDF parsing models running on Nebius.
Pleias describes synthetic training data generation and Nano model training with TractoAI. It adapted tokenized data tables, Nanotron and checkpoint saving, and used tractorun to deploy and coordinate distributed training automatically with fault tolerance.
They Said It Couldn’t Be Done5 Dec 2024 · Hugging Face · Web pagePleias describes synthetic training data generation and Nano model training with TractoAI. It adapted tokenized data tables, Nanotron and checkpoint saving, and used tractorun to deploy and coordinate distributed training automatically with fault tolerance.
PleIAs/Pleias-Nano · Hugging Face
19 Oct 2026 · huggingface.co · Web page
Pleias documents continuous pretraining of its Nano RAG model on TractoAI using Nebius AI’s ISEG GPU cluster and TractoAI’s Nanotron fork. The retrieval dataset contains about 45 billion tokens.
PleIAs/Pleias-Nano · Hugging Face19 Oct 2026 · huggingface.co · Web pagePleias documents continuous pretraining of its Nano RAG model on TractoAI using Nebius AI’s ISEG GPU cluster and TractoAI’s Nanotron fork. The retrieval dataset contains about 45 billion tokens.
Pleias Nano model card
18 Oct 2026 · Hugging Face · Web page
Pleias reports full Nano preview pretraining on TractoAI using Nebius AI’s ISEG cluster with 192 H100 GPUs for five days. Its schedule comprised 518,000 steps with a batch size of 1,024 and nearly five trillion tokens over three epochs. Estimated training emissions were four tonnes CO2eq.
Pleias Nano model card18 Oct 2026 · Hugging Face · Web pagePleias reports full Nano preview pretraining on TractoAI using Nebius AI’s ISEG cluster with 192 H100 GPUs for five days. Its schedule comprised 518,000 steps with a batch size of 1,024 and nearly five trillion tokens over three epochs. Estimated training emissions were four tonnes CO2eq.
Today pleias announces the release of Common Corpus 2. The largest fully open corpus for pretraining comes back better than ever: 2 trillion tokens with document-level licensing, provenance and… | Pierre-Carl Langlais
19 Oct 2026 · www.linkedin.com · Web page
Pierre-Carl Langlais states that Pleias processed 11 million scientific articles from OpenAlex, about 175 billion tokens, using PDF parsing models on Nebius for Common Corpus 2.
Today pleias announces the release of Common Corpus 2. The largest fully open corpus for pretraining comes back better than ever: 2 trillion tokens with document-level licensing, provenance and… | Pierre-Carl Langlais19 Oct 2026 · www.linkedin.com · Web pagePierre-Carl Langlais states that Pleias processed 11 million scientific articles from OpenAlex, about 175 billion tokens, using PDF parsing models on Nebius for Common Corpus 2.
Releasing the largest multilingual open pretraining dataset
13 Nov 2024 · Hugging Face · Web page
Pleias credits Nebius AI and Tracto AI among the organizations supporting Common Corpus storage and processing.
Releasing the largest multilingual open pretraining dataset13 Nov 2024 · Hugging Face · Web pagePleias credits Nebius AI and Tracto AI among the organizations supporting Common Corpus storage and processing.
Pleias is a French private AI company developing data infrastructure and open language models for enterprise teams, institutions, researchers and developers. Its Stratum product converts unstructured documents into structured data for AI agents and retrieval workflows, with privacy controls and on-premise deployment. Synth generates synthetic training data for specialized models and agents. Pleias also maintains Common Corpus, an openly licensed training dataset, and releases small language models and document-processing tools for tasks such as text extraction, OCR correction and source-grounded question answering.
Documented financing · private company
AmountUndisclosedMozilla Builders accelerator grant
Pleias releases an open Wikidata search agent and training traces
Pleias released an open search harness developed with AI-BRIDGES at the University of London. It lets language models query Wikidata through code and includes 10,235 search traces for training smaller models.
Pleias announces release of its Baguettotron model suite
Pleias researchers announced the public release of Baguettotron models under a permissive license alongside their SYNTH research paper. The suite includes dense reasoning models and a mixture-of-experts model trained using synthetic data.
Pleias develops AI training datasets, synthetic data tools, document processing infrastructure and open language models. · Infrastructure
How they use Nebius
Pleias pretrained its Nano 1.2B preview model on TractoAI using Nebius AI’s ISEG GPU cluster. Its model card reports 192 H100 GPUs over five days. The training code used TractoAI’s fork of Nanotron. The training schedule comprised 518,000 steps with a batch size of 1,024, covering nearly five trillion tokens over three epochs. Pleias estimated training emissions at four tonnes CO2eq. Pleias also continuously pretrained its Nano RAG variant on TractoAI using the ISEG cluster, with a retrieval dataset containing about 45 billion tokens.
Pleias converted tokenized training data into TractoAI tables, built a Nanotron dataset adapter and adapted file operations, including checkpoint saving. It used tractorun to deploy and coordinate distributed training automatically with fault tolerance. Pleias also generated synthetic RAG and instruction training data with larger models through a Map-Reduce based TractoAI method.
For Common Corpus, Pleias credited Nebius AI and Tracto AI among the organizations supporting storage and processing. Its release article describes a multilingual training dataset containing more than two trillion permissibly licensed tokens. For Common Corpus 2, Pleias processed 11 million scientific articles from OpenAlex, representing about 175 billion tokens, using PDF parsing models running on Nebius.
Pleias describes synthetic training data generation and Nano model training with TractoAI. It adapted tokenized data tables, Nanotron and checkpoint saving, and used tractorun to deploy and coordinate distributed training automatically with fault tolerance.
They Said It Couldn’t Be Done5 Dec 2024 · Hugging Face · Web pagePleias describes synthetic training data generation and Nano model training with TractoAI. It adapted tokenized data tables, Nanotron and checkpoint saving, and used tractorun to deploy and coordinate distributed training automatically with fault tolerance.
PleIAs/Pleias-Nano · Hugging Face
19 Oct 2026 · huggingface.co · Web page
Pleias documents continuous pretraining of its Nano RAG model on TractoAI using Nebius AI’s ISEG GPU cluster and TractoAI’s Nanotron fork. The retrieval dataset contains about 45 billion tokens.
PleIAs/Pleias-Nano · Hugging Face19 Oct 2026 · huggingface.co · Web pagePleias documents continuous pretraining of its Nano RAG model on TractoAI using Nebius AI’s ISEG GPU cluster and TractoAI’s Nanotron fork. The retrieval dataset contains about 45 billion tokens.
Pleias Nano model card
18 Oct 2026 · Hugging Face · Web page
Pleias reports full Nano preview pretraining on TractoAI using Nebius AI’s ISEG cluster with 192 H100 GPUs for five days. Its schedule comprised 518,000 steps with a batch size of 1,024 and nearly five trillion tokens over three epochs. Estimated training emissions were four tonnes CO2eq.
Pleias Nano model card18 Oct 2026 · Hugging Face · Web pagePleias reports full Nano preview pretraining on TractoAI using Nebius AI’s ISEG cluster with 192 H100 GPUs for five days. Its schedule comprised 518,000 steps with a batch size of 1,024 and nearly five trillion tokens over three epochs. Estimated training emissions were four tonnes CO2eq.
Today pleias announces the release of Common Corpus 2. The largest fully open corpus for pretraining comes back better than ever: 2 trillion tokens with document-level licensing, provenance and… | Pierre-Carl Langlais
19 Oct 2026 · www.linkedin.com · Web page
Pierre-Carl Langlais states that Pleias processed 11 million scientific articles from OpenAlex, about 175 billion tokens, using PDF parsing models on Nebius for Common Corpus 2.
Today pleias announces the release of Common Corpus 2. The largest fully open corpus for pretraining comes back better than ever: 2 trillion tokens with document-level licensing, provenance and… | Pierre-Carl Langlais19 Oct 2026 · www.linkedin.com · Web pagePierre-Carl Langlais states that Pleias processed 11 million scientific articles from OpenAlex, about 175 billion tokens, using PDF parsing models on Nebius for Common Corpus 2.
Releasing the largest multilingual open pretraining dataset
13 Nov 2024 · Hugging Face · Web page
Pleias credits Nebius AI and Tracto AI among the organizations supporting Common Corpus storage and processing.
Releasing the largest multilingual open pretraining dataset13 Nov 2024 · Hugging Face · Web pagePleias credits Nebius AI and Tracto AI among the organizations supporting Common Corpus storage and processing.
Pleias is a French private AI company developing data infrastructure and open language models for enterprise teams, institutions, researchers and developers. Its Stratum product converts unstructured documents into structured data for AI agents and retrieval workflows, with privacy controls and on-premise deployment. Synth generates synthetic training data for specialized models and agents. Pleias also maintains Common Corpus, an openly licensed training dataset, and releases small language models and document-processing tools for tasks such as text extraction, OCR correction and source-grounded question answering.
Documented financing · private company
AmountUndisclosedMozilla Builders accelerator grant
Pleias releases an open Wikidata search agent and training traces
Pleias released an open search harness developed with AI-BRIDGES at the University of London. It lets language models query Wikidata through code and includes 10,235 search traces for training smaller models.
Pleias announces release of its Baguettotron model suite
Pleias researchers announced the public release of Baguettotron models under a permissive license alongside their SYNTH research paper. The suite includes dense reasoning models and a mixture-of-experts model trained using synthetic data.
