# /// script
# requires-python = ">=3.10"
# dependencies = [
# "numpy",
# "torch==2.8.0",
# "kernels-benchmark-tools",
# "kernels",
# ]
#
# [tool.uv.sources]
# kernels-benchmark-tools = { path = "../../../../../tools", editable = true }
# ///
import torch
import sys
from kernels_benchmark_tools import KernelTypeEnum, run_benchmark
from kernels import get_kernel
# Load the causal conv1d kernel
causal_conv1d = get_kernel("kernels-community/causal-conv1d")
def hf_kernels_causal_conv1d(input_tensor, weight, bias):
return causal_conv1d.causal_conv1d_fn(input_tensor, weight, bias)
run_benchmark(
kernel_type=KernelTypeEnum.CAUSAL_CONV1D,
impl_name="hf_kernels_causal_conv1d",
impl_tags={"family": "hf-kernels", "backend": "cuda"},
impl_func=hf_kernels_causal_conv1d,
)