| 52 | } |
| 53 | |
| 54 | size_t kv_cache_bytes(const TargetCache & c) { |
| 55 | size_t n = 0; |
| 56 | for (auto * t : c.attn_k) if (t) n += ggml_nbytes(t); |
| 57 | for (auto * t : c.attn_v) if (t) n += ggml_nbytes(t); |
| 58 | return n; |
| 59 | } |
| 60 | |
| 61 | size_t vram_used_now() { |
| 62 | size_t free_b = 0, total_b = 0; |