fix

Author: sufubao

lightllm/common/fused_moe/moe_silu_and_mul.py

@@ -3,7 +3,7 @@
 import triton
 import triton.language as tl
 from .moe_silu_and_mul_config import MoeSiluAndMulKernelConfig
-from lightllm.common.triton_utils.autotuner import autotune, closest_pow_of_2
+from lightllm.common.triton_utils.autotuner import autotune
 
 
 @triton.jit
@@ -81,7 +81,7 @@ def _get_silu_and_mul_static_key(input: torch.Tensor, output: torch.Tensor):
     kernel_name="silu_and_mul_fwd:v1",
     configs_gen_func=_get_silu_and_mul_configs,
     static_key_func=_get_silu_and_mul_static_key,
-    run_key_func=lambda input: closest_pow_of_2(input.shape[0]),
+    run_key_func=lambda input: input.shape[0],
     mutates_args=["output"],
 )
 def silu_and_mul_fwd(input: torch.Tensor, output: torch.Tensor, run_config=None):

lightllm/common/fused_moe/moe_sum_reduce.py

@@ -4,7 +4,7 @@
 import triton.language as tl
 from .moe_sum_recude_config import MoeSumReduceKernelConfig
 from typing import Any, Callable, Dict, Optional, Tuple
-from lightllm.common.triton_utils.autotuner import autotune, closest_pow_of_2
+from lightllm.common.triton_utils.autotuner import autotune
 
 
 @triton.jit
@@ -66,7 +66,7 @@ def _get_moe_sum_reduce_configs():
     kernel_name="moe_sum_reduce:v1",
     configs_gen_func=_get_moe_sum_reduce_configs,
     static_key_func=_get_moe_sum_reduce_static_key,
-    run_key_func=lambda input: closest_pow_of_2(input.shape[0]),
+    run_key_func=lambda input: input.shape[0],
     mutates_args=["output"],
 )
 def moe_sum_reduce(input: torch.Tensor, output: torch.Tensor, run_config: Dict = None):

lightllm/common/quantization/triton_quant/fp8/fp8w8a8_block_gemm_kernel.py

@@ -7,7 +7,7 @@
 from functools import lru_cache
 from typing import Any, Dict, List, Optional, Tuple
 from triton import Config
-from lightllm.common.triton_utils.autotuner import autotune, closest_pow_of_2
+from lightllm.common.triton_utils.autotuner import autotune
 
 
 class Fp8BlockMMKernelConfig(KernelConfigs):

lightllm/models/deepseek2/triton_kernel/rotary_emb.py

@@ -3,7 +3,7 @@
 import triton
 import triton.language as tl
 import itertools
-from lightllm.common.triton_utils.autotuner import autotune, closest_pow_of_2
+from lightllm.common.triton_utils.autotuner import autotune
 
 
 @triton.jit