获取指定层的KV缓存 Args: layer_idx: 层索引 Returns: Tuple[torch.Tensor, torch.Tensor]: (key_cache, value_cache)
(self, layer_idx)
| 44 | return len(self.cache_kwargs) |
| 45 | |
| 46 | def get_kvcache(self, layer_idx): |
| 47 | """ |
| 48 | 获取指定层的KV缓存 |
| 49 | |
| 50 | Args: |
| 51 | layer_idx: 层索引 |
| 52 | |
| 53 | Returns: |
| 54 | Tuple[torch.Tensor, torch.Tensor]: (key_cache, value_cache) |
| 55 | """ |
| 56 | key_cache = self.key_cache[layer_idx] |
| 57 | value_cache = self.value_cache[layer_idx] |
| 58 | return key_cache, value_cache |
| 59 | |
| 60 | def get_router_kcache(self, layer_idx): |
| 61 | if layer_idx < len(self.router_key_cache): |