Brief IA
›

OVHcloud Unveils Olala, Launch Set for November 19

AI Tools·Tom Levy·

OVHcloud Unveils Olala, Launch Set for November 19

OVHcloud Unveils Olala, Launch Set for November 19
Key Takeaways
1OVHcloud plans to officially launch Olala mini 7A1B on November 19.
2Qwant is already using the model for its "flash" responses.
3A version 1.1 is expected before the end of the year with a context of 128,000 tokens.
Why it matters : OVHcloud is now developing its own models to meet its needs and those of its clients.

OVHcloud presents Olala mini 7A1B, its first large language model, with an official launch scheduled for November 19. Having become an artificial intelligence laboratory, the group names its model family Olala.

OVHcloud aims to reduce dependence on other labs

For OVHcloud, designing its own models aims to limit reliance on external suppliers. Its CEO, Octave Klaba, states that the company must be able to create models for its internal needs and those of its clients, in addition to the AI services already offered. He believes that OVHcloud cannot solely depend on the future sharing of open models by other laboratories.

The company plans to develop larger models and focuses its strategy on three areas: AI agents for businesses, cloud DevOps, and cybersecurity. A version 1.1 is expected by the end of the year, with a context window increased to 128,000 tokens, up from 32,000 currently, and improvements anticipated in mathematical reasoning. A version 2.0 is also on the roadmap.

La lettre du soirGratuit · 19 h

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Qwant is already using Olala for its responses

Qwant is already relying on Olala mini 7A1B to produce its "flash" responses. According to OVHcloud, this allows the search engine to reduce response time by 20% and cut AI costs by a factor of three compared to the previously used model.

OVHcloud emphasizes the speed and performance of its model across various types of hardware. The company indicates that a GPU with at least 8 GB of memory is sufficient to run it, achieving a throughput of 160 tokens per second on an RTX A4500. On a CPU alone, without a GPU, the reported throughput reaches 50 tokens per second.

Olala activates one billion parameters

Olala mini 7A1B uses a hybrid architecture of 7 billion parameters, of which one billion are activated for each request. This means that only a portion of the model is activated with each use. For comparison, Mistral Large 4 has 1 trillion parameters, with 52 billion active. No benchmark is available for Olala mini 7A1B at this stage, making it difficult to compare its performance with other models.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.