TileQ applies 2D-tiled low-rank quantization to MoE experts and fuses computations for up to 10x lower memory overhead and ~5% inference latency while keeping accuracy.
The selection involves indexing into a (B, M r) tensor using (B,K, r) indices—costing O(BKr) memory operations
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
-
TileQ: Efficient Low-Rank Quantization of Mixture-of-Experts with 2D Tiling
TileQ applies 2D-tiled low-rank quantization to MoE experts and fuses computations for up to 10x lower memory overhead and ~5% inference latency while keeping accuracy.