Gahow Wang d217f4fbd3 perf: spread flash bwd dK/dV atomics across all threads ...

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

2026-06-17 23:27:33 +08:00

..

attention.cu

autograd: batch dim for ops (flatten linears, batched attention)

2026-06-16 00:44:15 +08:00

cast.cu

cuda: bf16 cuBLAS GemmEx (16BF in/out, fp32 accum) + cast kernels

2026-06-16 14:14:39 +08:00

elementwise.cu

tensor: add scale elementwise CUDA kernel + FFI

2026-06-15 15:13:06 +08:00

flash_attention.cu

perf: spread flash bwd dK/dV atomics across all threads

2026-06-17 23:27:33 +08:00

gemm.cu

gemm: tiled F32 forward + transpose + backward (dA/dB)

2026-06-15 15:26:51 +08:00

model.cu

autograd: batch dim for ops (flatten linears, batched attention)

2026-06-16 00:44:15 +08:00

nn.cu

autograd: batch dim for ops (flatten linears, batched attention)

2026-06-16 00:44:15 +08:00

optim.cu

perf: GPU AdamW + grad-norm

2026-06-15 16:53:09 +08:00