andodeki/makepad
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
makepad/libs/ai/cuda
History
Exact
Admin a2aaa8f778 ai-body: the FP8 bias rides a column-broadcast add on the device 
The FP8 backbone mode added each linear's bias by downloading, tiling and
re-uploading it per call, which cost more than the FP8 GEMM saved. A
gpu_add_cols_broadcast op (CUDA kernel; host loop on the Metal tensor
backend) adds a cols-wide bias to every row on the device, and each linear
keeps its bias resident after the first upload.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
2026-09-02 00:05:03 +02:00
..
kernels ai-body: the FP8 bias rides a column-broadcast add on the device 2026-09-02 00:05:03 +02:00
src ai-body: the FP8 bias rides a column-broadcast add on the device 2026-09-02 00:05:03 +02:00
tests/data libs/ai: one AI stack, replacing libs/ggml, llama, mlx, cuda, tts, voice2 and pbr_paint 2026-08-23 01:34:35 +02:00
build.rs speech: one STT/TTS API on every platform, through the ai-hub 2026-09-01 23:32:35 +02:00
Cargo.toml libs/ai: one AI stack, replacing libs/ggml, llama, mlx, cuda, tts, voice2 and pbr_paint 2026-08-23 01:34:35 +02:00