What is Luminous?
Luminous Model Family
The Luminous series is a family of large language models. Large language models are powerful technological tools that can process and produce human text. They have acquired this ability by “reading” enormous amounts of human text data during training. Similar to someone reading a whole library and half of the internet, large language models have acquired a structural understanding of language and accumulated information about the world.
The Luminous family currently consists of three vanilla models, which differ in complexity and ability. They are, from the smallest to the largest, Luminous-base, Luminous-extended and Luminous-supreme. All Luminous models have been trained in the five most commonly spoken European languages: English, German, French, Italian and Spanish. Contrary to other large language models, the Luminous family posseses the ability to work with images as well. We call this multimodality. All three vanilla models are also available in an instruction-finetuned version, our Luminous-control models.
| Model | Description |
|---|---|
| Luminous-supreme | Luminous-supreme is the largest model in the Luminous model family and consequently the most capable one. It can solve all tasks the smaller models can solve as well and is especially well suited for creative text writing. |
| Luminous-extended | Luminous-extended is our second-largest model. It is well suited for tasks like information extraction and language simplification. It performs better on a wide range of tasks compared to Luminous-base and is faster and cheaper than Luminous-supreme. |
| Luminous-base | Luminous-base is our smallest model. That makes it the fastest and cheapest model in the Luminous model family. Therefore, this model is suited for applications where speed is important and costs should be low. It is very well suited for tasks like classification and labelling. |
| Luminous-control models | Our control models are versions of models optimized to follow instructions. This means that these models have been fine-tuned for a diverse set of tasks described by human-written instructions. One of their great advantages is that they have much better zero-shot performance compared to our vanilla models. That makes them easier to use and also leads to fewer tokens in the input, resulting in lower costs. We provide a control version of a model for all three model sizes: base, extended, and supreme. |
Model Capabilities, Methods and Services
All of our models can handle text as input. Next to that, Luminous-base and Luminous-extended can also handle multimodal input. This means they can process a combination of image and text input and output text. Read more about it here.
A method abstracts a specific interaction with a model for a low-level purpose. Luminous-Explore is our model for semantic embeddings based on Luminous-base.
A service provides value on top of a model or independent of a specific model. These services are designed for a certain task, such as question answering or summarization.