jgrusewski
1d9aa6e32d
feat(bf16): ml-core compiles clean — nvrtc removed, BF16 boundaries fixed
- Deleted cuda_compile.rs (nvrtc runtime compilation)
- Removed nvrtc feature from cudarc dependency
- Added f16 feature + half crate
- Stubbed nvrtc-dependent constructors (ReductionKernels, GpuAdamW, etc.)
- Fixed all BF16 boundary conversions (f32 host ↔ bf16 GPU)
- GpuTensor.from_host/to_host now convert at boundary
ml-core: 0 errors. ml: 49 errors remaining (Phase 3 Task 14).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>