[Re-land][jit kernel] Support per_token_group_quant_8bit jit kernel (#19140)

Co-authored-by: luoyuan.luo <luoyuan.luo@antgroup.com>
Co-authored-by: Mohammad Miadh Angkad <mangkad.bsdsba2027@aim.edu>
This commit is contained in:
Yuan Luo
2026-02-26 09:53:57 +08:00
committed by GitHub
parent beabaa8d37
commit 7c9e8e2def
6 changed files with 839 additions and 14 deletions

View File

@@ -10,7 +10,6 @@ import torch
if TYPE_CHECKING:
from tvm_ffi import Module
F = TypeVar("F", bound=Callable[..., Any])
@@ -76,7 +75,9 @@ class CPPArgList(list[str]):
CPP_DTYPE_MAP = {
torch.float: "fp32_t",
torch.float16: "fp16_t",
torch.float8_e4m3fn: "fp8_e4m3_t",
torch.bfloat16: "bf16_t",
torch.int8: "int8_t",
}