Search papers, labs, and topics across Lattice.
Affiliation:
2
0
3
Standard BF16 FlashAttention silently breaks a fundamental zero-sum symmetry in the softmax backward pass, corrupting query gradients by over 200% and triggering thousandfold gradient blowups late in pretraining.
K2-V2 redefines the open-source LLM landscape, rivaling closed-source giants in reasoning capabilities while providing full transparency into its training data and process.