Simple LRU cache for dequantized expert weights. Avoids re-dequantizing the same popular experts across tokens.
| 20 | /// Simple LRU cache for dequantized expert weights. |
| 21 | /// Avoids re-dequantizing the same popular experts across tokens. |
| 22 | struct ExpertCache { |
| 23 | entries: RwLock<HashMap<usize, ExpertWeights>>, |
| 24 | capacity: usize, |
| 25 | } |
| 26 | |
| 27 | /// Pre-computed tensor names for a single expert (avoids format! on hot path). |
| 28 | struct ExpertNames { |
nothing calls this directly
no outgoing calls
no test coverage detected