When will a language model with at least 100B parameters be open sourced including for commercial use?