Update speed FAQ

src/about.py

@@ -90,7 +90,9 @@ https://huggingface.co/datasets/silma-ai/arabic-broad-benchmark#scoring-rules
 
 *   Speed shows how fast the model was during testing, using the "words per second" metric.
 *   The score is calculated by dividing the number of words generated by the model during the entire test by the time taken (in seconds) for the model to complete testing.
-*   Please note that we use the same GPU (A100) and a batch size of 1 for all models (in the same size category) to ensure a fair comparison.
+*   Please note that we use the same GPU (A100) and a batch size of 1 for all Hugging Face models to ensure a fair comparison. Models above 15B are split across multiple GPUs.
+*   Each model should only be compared to other models in its size category.
+*   API or closed models can't be compared to open models, only to other API models, since they are not hosted on our infrastructure.
 
 ---