RTPurbo exploits intrinsic sparsity in full-attention LLMs to achieve near-lossless sparse inference after only a few hundred training steps via retrieval-head identification and a lightweight token indexer.
Aime 2024 dataset, 2024
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Full Attention Strikes Back: Transferring Full Attention into Sparse within Hundred Training Steps
RTPurbo exploits intrinsic sparsity in full-attention LLMs to achieve near-lossless sparse inference after only a few hundred training steps via retrieval-head identification and a lightweight token indexer.