[Bugfix] Fix triton import with local TritonPlaceholder (#17446)

Signed-off-by: Mengqing Cao <cmq0113@163.com>
This commit is contained in:
Mengqing Cao
2025-05-06 17:53:09 +08:00
committed by GitHub
parent 05e1f96419
commit f9bc5a0693
30 changed files with 165 additions and 75 deletions
@@ -6,8 +6,6 @@ import os
from typing import Any, Callable, Dict, List, Optional, Tuple
import torch
import triton
import triton.language as tl
import vllm.envs as envs
from vllm import _custom_ops as ops
@@ -21,6 +19,7 @@ from vllm.model_executor.layers.quantization.utils.fp8_utils import (
from vllm.model_executor.layers.quantization.utils.int8_utils import (
per_token_group_quant_int8, per_token_quant_int8)
from vllm.platforms import current_platform
from vllm.triton_utils import tl, triton
from vllm.utils import direct_register_custom_op
from .rocm_aiter_fused_moe import is_rocm_aiter_moe_enabled
@@ -2,11 +2,10 @@
from typing import Optional, Tuple
import torch
import triton
import triton.language as tl
import vllm.envs as envs
from vllm import _custom_ops as ops
from vllm.triton_utils import tl, triton
from vllm.utils import round_up
+2 -2
View File
@@ -1,9 +1,9 @@
# SPDX-License-Identifier: Apache-2.0
import torch
import triton
import triton.language as tl
from einops import rearrange
from vllm.triton_utils import tl, triton
@triton.jit
def _fwd_diag_kernel(Q, K, V, Out, S, b: tl.constexpr, h: tl.constexpr, n,
@@ -4,13 +4,11 @@
# Adapted from https://github.com/state-spaces/mamba/blob/v2.2.4/mamba_ssm/ops/triton/selective_state_update.py
import torch
import triton
import triton.language as tl
from packaging import version
from vllm import _custom_ops as ops
from vllm.attention.backends.utils import PAD_SLOT_ID
from vllm.triton_utils import HAS_TRITON
from vllm.triton_utils import HAS_TRITON, tl, triton
TRITON3 = HAS_TRITON and (version.parse(triton.__version__)
>= version.parse("3.0.0"))
@@ -8,8 +8,8 @@
import math
import torch
import triton
import triton.language as tl
from vllm.triton_utils import tl, triton
@triton.autotune(
@@ -6,10 +6,10 @@
# ruff: noqa: E501,SIM102
import torch
import triton
import triton.language as tl
from packaging import version
from vllm.triton_utils import tl, triton
TRITON_22 = version.parse(triton.__version__) >= version.parse('2.2.0')
@@ -8,8 +8,8 @@
import math
import torch
import triton
import triton.language as tl
from vllm.triton_utils import tl, triton
from .mamba_ssm import softplus
@@ -6,10 +6,11 @@
# ruff: noqa: E501
import torch
import triton
from einops import rearrange
from packaging import version
from vllm.triton_utils import triton
from .ssd_bmm import _bmm_chunk_fwd
from .ssd_chunk_scan import _chunk_scan_fwd
from .ssd_chunk_state import (_chunk_cumsum_fwd, _chunk_state_fwd,
@@ -6,8 +6,8 @@
# ruff: noqa: E501
import torch
import triton
import triton.language as tl
from vllm.triton_utils import tl, triton
@triton.autotune(
@@ -1,8 +1,8 @@
# SPDX-License-Identifier: Apache-2.0
import torch
import triton
import triton.language as tl
from vllm.triton_utils import tl, triton
AWQ_TRITON_SUPPORTED_GROUP_SIZES = [-1, 32, 64, 128]
@@ -3,8 +3,8 @@
from typing import Optional, Type
import torch
import triton
import triton.language as tl
from vllm.triton_utils import tl, triton
def is_weak_contiguous(x: torch.Tensor):
@@ -7,8 +7,6 @@ import os
from typing import Any, Dict, List, Optional, Tuple, Union
import torch
import triton
import triton.language as tl
from vllm import _custom_ops as ops
from vllm.logger import init_logger
@@ -17,6 +15,7 @@ from vllm.model_executor.layers.quantization.utils.quant_utils import (
from vllm.model_executor.layers.quantization.utils.w8a8_utils import (
CUTLASS_BLOCK_FP8_SUPPORTED)
from vllm.platforms import current_platform
from vllm.triton_utils import tl, triton
from vllm.utils import direct_register_custom_op
logger = init_logger(__name__)
@@ -8,10 +8,9 @@ import os
from typing import Any, Dict, List, Optional, Tuple
import torch
import triton
import triton.language as tl
from vllm.platforms import current_platform
from vllm.triton_utils import tl, triton
logger = logging.getLogger(__name__)