| .. |
|
llm
|
work: Relax JIT campaign batch 2026-10-23 (owner/PGO hang fix, JIT memory + run monitor fix, rustc suite 99.70%, Flow OSS app, skins OSS, Stage logs, Rogier + makepad PR syncs)
|
2026-10-23 15:18:28 +02:00 |
|
affine.cu
|
libs/ai: one AI stack, replacing libs/ggml, llama, mlx, cuda, tts, voice2 and pbr_paint
|
2026-08-23 01:34:35 +02:00 |
|
diffusion_ops.cu
|
ai: the AI hub gives way to the UI, smaller models, vision and speech
|
2026-10-13 12:26:21 +02:00 |
|
gated_delta_net.cu
|
libs/ai: one AI stack, replacing libs/ggml, llama, mlx, cuda, tts, voice2 and pbr_paint
|
2026-08-23 01:34:35 +02:00 |
|
gather.cu
|
ai: the AI hub gives way to the UI, smaller models, vision and speech
|
2026-10-13 12:26:21 +02:00 |
|
kokoro_ops.cu
|
ai: the AI hub gives way to the UI, smaller models, vision and speech
|
2026-10-13 12:26:21 +02:00 |
|
kquants.cu
|
ai: the AI hub gives way to the UI, smaller models, vision and speech
|
2026-10-13 12:26:21 +02:00 |
|
launch_profile.cu
|
ai: the AI hub gives way to the UI, smaller models, vision and speech
|
2026-10-13 12:26:21 +02:00 |
|
launch_profile.cuh
|
ai: the AI hub gives way to the UI, smaller models, vision and speech
|
2026-10-13 12:26:21 +02:00 |
|
nvfp4.cu
|
libs/ai: one AI stack, replacing libs/ggml, llama, mlx, cuda, tts, voice2 and pbr_paint
|
2026-08-23 01:34:35 +02:00 |
|
nvfp4_mmq.cu
|
libs/ai: one AI stack, replacing libs/ggml, llama, mlx, cuda, tts, voice2 and pbr_paint
|
2026-08-23 01:34:35 +02:00 |
|
ops.cu
|
ai: the AI hub gives way to the UI, smaller models, vision and speech
|
2026-10-13 12:26:21 +02:00 |
|
paint_extras.cu
|
libs/ai: one AI stack, replacing libs/ggml, llama, mlx, cuda, tts, voice2 and pbr_paint
|
2026-08-23 01:34:35 +02:00 |
|
pixal.cu
|
libs: ai hub/models/cuda, speech, chat_ui
|
2026-09-15 13:40:31 +02:00 |
|
q8.cu
|
ai: the AI hub gives way to the UI, smaller models, vision and speech
|
2026-10-13 12:26:21 +02:00 |
|
q8_dense.cu
|
ai: the AI hub gives way to the UI, smaller models, vision and speech
|
2026-10-13 12:26:21 +02:00 |
|
rife.cu
|
libs/ai: one AI stack, replacing libs/ggml, llama, mlx, cuda, tts, voice2 and pbr_paint
|
2026-08-23 01:34:35 +02:00 |
|
roformer.cu
|
ai backbone: the hub era — makepad_ai deleted, every backend is a hub pipe; machine residency elections, job leases, ETA placement; the store only stores; creator pipelines run in the app (aicore)
|
2026-09-01 16:46:31 +02:00 |
|
sampling.cu
|
ai: the AI hub gives way to the UI, smaller models, vision and speech
|
2026-10-13 12:26:21 +02:00 |
|
sequence_ops.cu
|
ai: the AI hub gives way to the UI, smaller models, vision and speech
|
2026-10-13 12:26:21 +02:00 |
|
splat.cu
|
libs/ai: one AI stack, replacing libs/ggml, llama, mlx, cuda, tts, voice2 and pbr_paint
|
2026-08-23 01:34:35 +02:00 |
|
ssm_conv.cu
|
libs/ai: one AI stack, replacing libs/ggml, llama, mlx, cuda, tts, voice2 and pbr_paint
|
2026-08-23 01:34:35 +02:00 |
|
train.cu
|
ai cuda: a training surface -- kernels/train.cu and src/train.rs give the forward and backward of a small op set (activations, row/column broadcasts, column copies, row gather/scatter, im2col/col2im, depthwise conv, LayerNorm, RoPE, head split/merge, masked softmax), masked L1/MSE/BCE losses, the STFT family (a shared-memory batched FFT, framing, Hermitian fill, overlap-add), a harmonic source, AdamW with global-norm clipping and EMA, counter-based normal noise, and monotonic alignment search; device buffers come from a size-class pool, uploads go through two pinned arenas so a training step never waits on a copy, and GEMMs are row-major strided-batched cuBLAS on TF32 tensor cores (switchable to exact f32). Every *_bwd kernel accumulates. Used by the singing model's device backend (libs/ai/models/sing, nn_gpu.rs). Checked: sm_120a build on node 165; sing_train check -- every loss and parameter gradient of both singing models matches the CPU reference to 3.9e-4 with exact f32 GEMMs (1.9e-2 with TF32), and the device harmonic source matches the host one to 2.9e-4 of its peak.
|
2026-09-30 22:51:29 +02:00 |
|
whisper_enc.cu
|
ai: the AI hub gives way to the UI, smaller models, vision and speech
|
2026-10-13 12:26:21 +02:00 |
