As some other renderers do, we share a single GPU buffer for all vertex data. Instead, we use a simple allocator which manages this contigous buffer automatically.

I'm not sure what this part is supposed to say, but it doesn't look right. "Instead" usually follows differences, not similarities.