MCPcopy Create free account
hub / github.com/RightNow-AI/autokernel / kernel_fn

Function kernel_fn

kernels/cuda/cross_entropy.py:151–162  ·  view source on GitHub ↗

Entry point called by bench.py. Must match reference.cross_entropy_ref signature.

(logits: torch.Tensor, targets: torch.Tensor)

Source from the content-addressed store, hash-verified

149
150
151def kernel_fn(logits: torch.Tensor, targets: torch.Tensor) -> torch.Tensor:
152 """Entry point called by bench.py. Must match reference.cross_entropy_ref signature."""
153 assert logits.is_cuda and targets.is_cuda
154
155 if logits.dtype != torch.float16:
156 logits = logits.to(torch.float16)
157
158 mod = _get_module()
159 loss = mod.cross_entropy_cuda(logits, targets)
160
161 # Loss is always returned as float32 (scalar loss, matches F.cross_entropy)
162 return loss.to(torch.float32)

Callers

nothing calls this directly

Calls 1

_get_moduleFunction · 0.70

Tested by

no test coverage detected