> For the complete documentation index, see [llms.txt](https://trust.memori.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://trust.memori.ai/en/legal-documents/gpai-general-purpose-ai-models.md).

# GPAI: General-Purpose AI Models

Memori s.r.l. compliance with Regulation (EU) 2024/1689 — AI Act, Chapter V

Chapter V of Regulation (EU) 2024/1689, known as the AI Act, introduces a specific regulatory framework for **general-purpose AI models (GPAI)**: systems trained at scale, capable of performing a wide range of distinct tasks and integrable into many downstream applications. Memori s.r.l. falls within the scope of this framework as the **provider of the MemoriNLP model**, placed on the Union market on 22 December 2021.

***

### What is meant by a GPAI model

* **Significant generality:** ability to perform different tasks in heterogeneous application contexts, not limited to a single domain or function
* **Computational threshold:** training with computing power above 10²⁵ FLOP — the quantitative criterion set by Art. 51 of the AI Act for classification
* **Downstream integration:** the model can be incorporated into AI systems developed by third parties (downstream providers), amplifying its impact on the market
* **Systemic risk:** Art. 51 of the AI Act establishes that GPAI models trained with computing power above 10²⁵ FLOP are presumed to have high systemic impact and are subject to the enhanced obligations of Art. 55. MemoriNLP is well below this threshold.

***

### How Memori is aligned

MemoriNLP is a GPAI-type NLP model trained exclusively on encyclopedic Wikipedia content (IT, EN, FR) under the CC BY-SA 3.0 license. Memori has adopted a structured compliance approach that covers all the obligations set out in Art. 53 and, where applicable, Art. 55 of the AI Act.

* **Training data transparency:** public documentation of the sources, linguistic characteristics and anti-bias measures adopted
* **Copyright compliance:** exclusive use of datasets with licenses compatible with text and data mining under Directive (EU) 2019/790
* **User data protection:** user data remains the property of the customer — Memori does not collect or use interaction data to train, modify or update the model
* **Information for downstream providers:** technical documentation and instructions for use made available to enable responsible integration of the model
* **Monitoring and incident reporting:** internal procedures for detecting and reporting serious incidents in line with the obligations of the AI Act

***

### Available GPAI documentation

Memori produces and keeps up to date a set of GPAI documents structured for different audiences — from the European AI Office to Data Protection Officers, up to the general public.

{% tabs %}
{% tab title="AIO - AI OFFICE" %}
[AIO — AI Office ](/en/legal-documents/gpai-general-purpose-ai-models/gpai-document-aio.md)\
Complete technical documentation drawn up under Measure 1.1 of the GPAI Code of Practice. It contains the architecture, training parameters, computational resources, benchmarks and the Safety & Security Framework.

\
Restricted — AI Office only
{% endtab %}

{% tab title="NCAs - National Authorities" %}
NCAs — National Competent Authorities\
Structured version of the technical documentation intended for the National Competent Authorities (NCAs). It includes the technical and governance elements needed for oversight and supervisory activities.\
\
Restricted — On formal request only
{% endtab %}

{% tab title="DPs - Providers" %}
[DPs — Downstream Providers](/en/legal-documents/gpai-general-purpose-ai-models/gpai-document-dps.md)\
Information for providers who integrate MemoriNLP into their own applications. It describes capabilities, limitations, permitted use cases and instructions for responsible integration.\
\
Available to registered partners
{% endtab %}

{% tab title="Public Summary of Training Content" %}
[Public Summary of Training Content](/en/legal-documents/gpai-general-purpose-ai-models/public-summary-of-training-content.md)

Public summary of the training data, drawn up according to the official template of the European Commission (v2.0), pursuant to Art. 53(1)(d) of the AI Act.
{% endtab %}
{% endtabs %}