Upload folder using huggingface_hub

torch-ext/triton_kernels/matmul_ogs_details/opt_flags.py

@@ -2,7 +2,7 @@
 # fmt: off
 from dataclasses import dataclass
 import triton
-from triton_kernels.target_info import get_cdna_version
+from ..target_info import get_cdna_version
 import torch
 from .opt_flags_details import opt_flags_amd, opt_flags_nvidia
 

torch-ext/triton_kernels/matmul_ogs_details/opt_flags_details/opt_flags_amd.py

@@ -1,7 +1,7 @@
 import torch
 import triton
-from triton_kernels.target_info import get_cdna_version
-from triton_kernels.tensor import bitwidth
+from ...target_info import get_cdna_version
+from ...tensor import bitwidth
 
 
 def compute_block_nk(n, block_m, grid_m, num_xcds, lhs_dtype, rhs_dtype, precision_config):

torch-ext/triton_kernels/matmul_ogs_details/opt_flags_details/opt_flags_nvidia.py

@@ -1,9 +1,9 @@
 import torch
 import triton
 from triton_kernels import target_info
-from triton_kernels.tensor import get_layout, bitwidth, FP4
-from triton_kernels.tensor_details.layout import HopperMXScaleLayout
-from triton_kernels.numerics_details.mxfp_details._downcast_to_mxfp import MXFP_BLOCK_SIZE
+from ...tensor import get_layout, bitwidth, FP4
+from ...tensor_details.layout import HopperMXScaleLayout
+from ...numerics_details.mxfp_details._downcast_to_mxfp import MXFP_BLOCK_SIZE
 
 
 def compute_grid_size(routing_data, m, n, block_m, block_n):