So open source dynamics only emerge when these two conditions are met. And in terms of stable diffusion, I think these are met earlier. And then when we participate in hugging face, quantizing models, fine tuning LoRA, I would say around june, july, the same dynamic emerged in hugging face around the language model space on llama weights. And later on, of course, many others, like Mixtral, became essentially the new mainline. But the dynamic is already there.