From 4c997310f530deb576c7c38dddced93ee21d30dd Mon Sep 17 00:00:00 2001 From: Xiaoyu Zhang <1182563586@qq.com> Date: Tue, 14 Jul 2026 08:41:23 +0800 Subject: [PATCH] [Kernel] Hotfix: update sgl-kernel imports of relocated fp8_kernel (RFC #29630 #30784) (#31089) Co-authored-by: Claude Opus 4.8 (1M context) --- sgl-kernel/benchmark/bench_fp8_blockwise_gemm.py | 2 +- sgl-kernel/benchmark/bench_per_token_group_quant_8bit.py | 6 +++--- sgl-kernel/tests/test_per_token_group_quant_8bit.py | 4 ++-- 3 files changed, 6 insertions(+), 6 deletions(-) diff --git a/sgl-kernel/benchmark/bench_fp8_blockwise_gemm.py b/sgl-kernel/benchmark/bench_fp8_blockwise_gemm.py index 29bd0a80a..f05687261 100644 --- a/sgl-kernel/benchmark/bench_fp8_blockwise_gemm.py +++ b/sgl-kernel/benchmark/bench_fp8_blockwise_gemm.py @@ -20,7 +20,7 @@ except ImportError: vllm_scaled_mm = None VLLM_AVAILABLE = False -from sglang.srt.layers.quantization.fp8_kernel import ( +from sglang.kernels.ops.quantization.fp8_kernel import ( w8a8_block_fp8_matmul_triton as w8a8_block_fp8_matmul, ) diff --git a/sgl-kernel/benchmark/bench_per_token_group_quant_8bit.py b/sgl-kernel/benchmark/bench_per_token_group_quant_8bit.py index e4a8e7bbd..58a3c6433 100644 --- a/sgl-kernel/benchmark/bench_per_token_group_quant_8bit.py +++ b/sgl-kernel/benchmark/bench_per_token_group_quant_8bit.py @@ -8,13 +8,13 @@ import torch import triton from sgl_kernel.test_utils import create_per_token_group_quant_test_data -from sglang.srt.layers.quantization.fp8_kernel import ( +from sglang.kernels.ops.quantization.fp8_kernel import ( create_per_token_group_quant_fp8_output_scale, ) -from sglang.srt.layers.quantization.fp8_kernel import ( +from sglang.kernels.ops.quantization.fp8_kernel import ( per_token_group_quant_8bit as triton_per_token_group_quant_8bit, ) -from sglang.srt.layers.quantization.fp8_kernel import ( +from sglang.kernels.ops.quantization.fp8_kernel import ( sglang_per_token_group_quant_8bit, ) from sglang.srt.utils import is_hip diff --git a/sgl-kernel/tests/test_per_token_group_quant_8bit.py b/sgl-kernel/tests/test_per_token_group_quant_8bit.py index 24ebeb877..38699314a 100644 --- a/sgl-kernel/tests/test_per_token_group_quant_8bit.py +++ b/sgl-kernel/tests/test_per_token_group_quant_8bit.py @@ -11,10 +11,10 @@ from sgl_kernel.test_utils import ( create_per_token_group_quant_test_data, ) -from sglang.srt.layers.quantization.fp8_kernel import ( +from sglang.kernels.ops.quantization.fp8_kernel import ( per_token_group_quant_8bit as triton_per_token_group_quant_8bit, ) -from sglang.srt.layers.quantization.fp8_kernel import ( +from sglang.kernels.ops.quantization.fp8_kernel import ( sglang_per_token_group_quant_8bit, ) from sglang.srt.utils import get_bool_env_var, is_hip