Efficient LLM Training and Serving with Heterogeneous Context Sharding among Attention Heads

Re-assign community