Utiliser un RAG est très efficace.
Le fine-tuning n'est pas souvent la solution.
https://www.tidepool.so/blog/why-you-probably-dont-need-to-fine-tune-an-llm
https://arxiv.org/abs/1706.03762
Epoch, ‘Parameter, Compute and Data Trends in Machine Learning’. Published online at epochai.org. Retrieved from: https://epochai.org/data/epochdb/visualization
https://blogs.nvidia.com/blog/tensorfloat-32-precision-format/
https://intellabs.github.io/distiller/algo_quantization.html