2 Commits

Include renames

Author	SHA1	Message	Date
Kristian MyramandClaude Sonnet 5.5	b7c78065db	S2.61: cache the sine positions, pool without permutes ... Sine position encodings depend only on the sizes: built once (150 ms of a frame). do_pool takes the maximum over grouped cells instead of permute + max_pool2d + permute (tested equal, odd sizes included). Encode 2.6 s -> 2.4 s per frame in release on 16 threads (3.4 s at the start of the speed work); reference tests unchanged. Chunked global attention was tried and did not help. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>	2026-10-30 08:43:31 +01:00
Kristian MyramandClaude Sonnet 5.5	aab00e43fe	S2.60: parallel exact GELU: encode 3.4 s -> 2.6 s per frame ... Candle's gelu_erf runs on one thread and was a third of the Hiera block time. A rayon CustomOp1 calls the same erf per element (bit-identical, tested). All reference tests unchanged. op_bench example times the block's operations. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>	2026-10-30 08:43:31 +01:00