Original release notes
metal: implement DSv4 Lightning Indexer (#25893)
- metal: implement F16 Lightning Indexer
- Implement GGML_OP_LIGHTNING_INDEXER for 128-dimensional, 64-head inputs
with F32 queries and weights plus F16 keys and masks.
- Add tiled and tail kernels and test KV lengths around 8- and 64-element
boundaries.
llama-bench (--mmap 1, -fa 1, -p 512, -n 128; d=0/10k/20k/30k):
Before:
- pp512: 153.73 0.87 t/s
- tg128: 8.91 0.04 t/s
- pp512 @ d10000: 73.90 0.39 t/s
- tg128 @ d10000: 8.66 0.03 t/s
- pp512 @ d20000: 45.83 0.18 t/s
- tg128 @ d20000: 8.26 0.03 t/s
- pp512 @ d30000: 33.40 0.21 t/s
- tg128 @ d30000: 7.94 0.01 t/s
After:
- pp512: 155.19 0.91 t/s
- tg128: 8.95 0.04 t/s
- pp512 @ d10000: 86.95 0.69 t/s
- tg128 @ d10000: 9.00 0.05 t/s
- pp512 @ d20000: 62.01 0.45 t/s
- tg128 @ d20000: 8.68 0.04 t/s
- pp512 @ d30000: 49.18 0.33 t/s
- tg128 @ d30000: 8.60 0.02 t/s
Assisted-by: Codex
- metal: stage Lightning Indexer K tiles
- Stage and dequantize K in F16 threadgroup memory before simdgroup matrix loads.
- Zero-fill partial tiles and guard stores so all KV segments use the same numerical path.
- Support F32, F16, BF16, Q4_0, Q4_1, Q5_0, Q5_1, and Q8_0 K caches.
llama-bench (--mmap 1, -fa on, -p 512, -n 128; d=0/10k/20k):
- pp512: 160.38 +/- 1.01 t/s
- tg128: 9.08 +/- 0.03 t/s
- pp512 @ d10000: 88.37 +/- 0.46 t/s
- tg128 @ d10000: 9.07 +/- 0.04 t/s
- pp512 @ d20000: 62.53 +/- 0.46 t/s
- tg128 @ d20000: 8.84 +/- 0.03 t/s
Assisted-by: Codex
- dedup Lightning Indexer constants, fix flaky test
- cont : fix whitespace
---------
Co-authored-by: forforever73 Co-authored-by: Georgi Gerganov
Website: -
UI:
