CuBridge adapts expert CUDA attention kernels via LLM-driven lift-transfer-lower to produce correct, high-performance implementations for new variants across GPUs.
FlashMask (Wang et al., 2025) introduces a structural sparse mask representation, allowing it to support a wider range of mask variants
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
CuBridge: An LLM-Based Framework for Understanding and Reconstructing High-Performance Attention Kernels
CuBridge adapts expert CUDA attention kernels via LLM-driven lift-transfer-lower to produce correct, high-performance implementations for new variants across GPUs.