vimarsh/repro-token-sparse-attention-long-context-interleaved-selection

Repro - Token Sparse Attention: Efficient Long-Context Inference with Interleaved Token Selection

An open experiment logbook, published with Trackio.