Kristian MyramandClaude Sonnet 5.5 ee3498b24d S2.63: GPU memory attention and memory encoder; one shared device; CPU fallback ...

Memory attention (1.8e-5 vs PyTorch) and the memory encoder (2.6e-6) as wgpu kernels,
convolution and layer-norm+GELU kernels, VideoModel::load_auto picks the GPU when
there is one (EFFECTCRAFT_SAM2_DEVICE overrides) and falls back to the CPU on any GPU
error; disable_gpu switches it off mid-clip. One process-wide device (several at once
crashed the Vulkan loader). A tracked frame: 4.3 s CPU, 0.7 s on the RX 7900 GRE.
Clips tracked on the GPU match Sam2VideoModel (worst logit error 0.0032).

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

2026-10-30 08:43:31 +01:00

..

gpu_matmul.rs

S2.61: parallel bias add fused with GELU; GPU matmul spike

2026-10-30 08:43:31 +01:00

op_bench.rs

S2.61: cache the sine positions, pool without permutes

2026-10-30 08:43:31 +01:00

sam2_bench.rs

S2.63: GPU memory attention and memory encoder; one shared device; CPU fallback

2026-10-30 08:43:31 +01:00