[Custom Ops] Add functional + out variant for scaled_fp4_quant (#34389)

Signed-off-by: tianrengao <terrygao87@gmail.com>
This commit is contained in:
Terry Gao
2026-03-16 18:51:46 -04:00
committed by GitHub
parent 7961486a9b
commit 3e6a1e1686
12 changed files with 213 additions and 44 deletions
@@ -63,7 +63,7 @@ QUANT_OPS: dict[QuantKey, OpOverload] = {
kFp8DynamicTokenSym: torch.ops._C.dynamic_per_token_scaled_fp8_quant.default, # noqa: E501
}
if current_platform.is_cuda() and hasattr(torch.ops._C, "scaled_fp4_quant"):
QUANT_OPS[kNvfp4Dynamic] = torch.ops._C.scaled_fp4_quant.default
QUANT_OPS[kNvfp4Dynamic] = torch.ops._C.scaled_fp4_quant.out
if current_platform.is_cuda():
QUANT_OPS[kFp8Dynamic128Sym] = torch.ops._C.per_token_group_fp8_quant.default # noqa: E501
QUANT_OPS[kFp8Dynamic64Sym] = torch.ops._C.per_token_group_fp8_quant.default # noqa: E501