feat: native bf16 GPU elementwise + AdamW kernels (ADR 075 L4)#152

Merged

dndungu merged 9 commits into

mainfrom

feat/bf16-gpu-elementwise-adamw

Jun 16, 2026

Commits on Jun 16, 2026

feat(kernels): bf16 GPU elementwise + AdamW CUDA kernels (ADR 075 L4)
dndungu
committed
feat(gpuapi): bf16 kernel methods on KernelRunner (ADR 075 L4)
dndungu
committed
feat(compute): native bf16 GPU elementwise + AdamW dispatch (ADR 075 L4)
dndungu
committed
test(spark): GB10 bf16 parity verify manifest (ADR 075 L4)
dndungu
committed
feat(kernels): bf16 forward-only fused norm CUDA kernels (ADR 075 L4)
dndungu
committed
feat(gpuapi): bf16 fused norm methods on KernelRunner (ADR 075 L4)
dndungu
committed
feat(compute): bf16 dispatch + parity tests for fused norm kernels (ADR 075 L4)
dndungu
committed
test(spark): use emptyDir for bf16-verify /out (hostPath dir absent on DGX)
dndungu
committed
test(compute): bf16 GEMM parity tolerance 1e-3 -> 5e-3 (pre-existing, GB10-exposed)
dndungu
committed