[Kernel] Remove unused fake implementation (#55535)

Signed-off-by: Jee Jee Li <jeejeelee@inferact.ai>
This commit is contained in:
Jee Jee Li
2026-09-06 17:54:32 -07:00
committed by GitHub
parent 4df80187b5
commit 199cb9b964
39 changed files with 7 additions and 956 deletions
@@ -293,24 +293,6 @@ if flashinfer_comm is not None:
or num_tokens > PDL_ADVANCE_LAUNCH_TOKENS,
)
def call_trtllm_fused_allreduce_norm_fake(
allreduce_in: torch.Tensor,
residual: torch.Tensor,
rms_gamma: torch.Tensor,
rms_eps: float,
world_size: int,
launch_with_pdl: bool,
fp32_acc: bool,
max_token_num: int,
pattern_code: int,
norm_out: torch.Tensor | None = None,
quant_out: torch.Tensor | None = None,
scale_out: torch.Tensor | None = None,
scale_factor: torch.Tensor | None = None,
weight_bias: float = 0.0,
) -> None:
pass
direct_register_custom_op(
op_name="flashinfer_trtllm_fused_allreduce_norm",
op_func=call_trtllm_fused_allreduce_norm,
@@ -321,7 +303,6 @@ if flashinfer_comm is not None:
"quant_out",
"scale_out",
],
fake_impl=call_trtllm_fused_allreduce_norm_fake,
)
flashinfer_trtllm_fused_allreduce_norm = (
torch.ops.vllm.flashinfer_trtllm_fused_allreduce_norm.default