Enabling High-Sparsity Foundational Llama Models with Efficient Pretraining and Deployment

MoE
SyDa
Re-assign community