c5dfa8d645
### What this PR does / why we need it?
This PR adds split_qkv_rmsnorm_mrope kernel with interleaved for qwen3.5
and qwen3-vl to improve performance.
### Does this PR introduce _any_ user-facing change?
Does not.
### How to use?
```python
real_q, real_k, real_v, real_gate = torch.ops.vllm.triton_split_qkv_rmsnorm_mrope(
qkv=qkv,
q_weight=q_weight,
k_weight=k_weight,
cos_sin=cos_sin,
num_q_heads=num_q_heads,
num_kv_heads=num_kv_heads,
head_size=head_size,
eps=eps,
mrope_section=mrope_section,
is_interleaved=is_interleaved,
rope_dim=rope_dim,
has_gate=has_gate,
)
```
### How was this patch tested?
- vLLM version: v0.16.0
- Accuracy test script：
```shell
pytest tests/e2e/nightly/single_node/ops/singlecard_ops/triton/test_split_qkv_rmsnorm_mrope.py
```
---------
Signed-off-by: Fager <865071616@qq.com>
Signed-off-by: Fager10086 <77871921+Fager10086@users.noreply.github.com>
Signed-off-by: fager <865071616@qq.com>
53 lines
2.3 KiB
Python
53 lines
2.3 KiB
Python
#
|
|
# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
# This file is a part of the vllm-ascend project.
|
|
#
|
|
|
|
import torch
|
|
from vllm.triton_utils import HAS_TRITON
|
|
|
|
import vllm_ascend.ops.fused_moe.fused_moe # noqa
|
|
import vllm_ascend.ops.layernorm # noqa
|
|
import vllm_ascend.ops.register_custom_ops # noqa
|
|
|
|
if HAS_TRITON:
|
|
import vllm_ascend.ops.triton.linearnorm.split_qkv_rmsnorm_rope # noqa
|
|
import vllm_ascend.ops.triton.linearnorm.split_qkv_rmsnorm_mrope
|
|
|
|
import vllm_ascend.ops.vocab_parallel_embedding # noqa
|
|
from vllm_ascend.ops.activation import AscendQuickGELU, AscendSiluAndMul
|
|
from vllm_ascend.ops.rotary_embedding import AscendDeepseekScalingRotaryEmbedding, AscendRotaryEmbedding
|
|
|
|
|
|
class dummyFusionOp:
|
|
default = None
|
|
|
|
def __init__(self, name=""):
|
|
self.name = name
|
|
|
|
|
|
def register_dummy_fusion_op() -> None:
|
|
torch.ops._C_ascend.rms_norm = dummyFusionOp(name="rms_norm")
|
|
torch.ops._C_ascend.fused_add_rms_norm = dummyFusionOp(name="fused_add_rms_norm")
|
|
torch.ops._C_ascend.static_scaled_fp8_quant = dummyFusionOp(name="static_scaled_fp8_quant")
|
|
torch.ops._C_ascend.dynamic_scaled_fp8_quant = dummyFusionOp(name="dynamic_scaled_fp8_quant")
|
|
torch.ops._C_ascend.dynamic_per_token_scaled_fp8_quant = dummyFusionOp(name="dynamic_per_token_scaled_fp8_quant")
|
|
torch.ops._C_ascend.rms_norm_static_fp8_quant = dummyFusionOp(name="rms_norm_static_fp8_quant")
|
|
torch.ops._C_ascend.fused_add_rms_norm_static_fp8_quant = dummyFusionOp(name="fused_add_rms_norm_static_fp8_quant")
|
|
torch.ops._C_ascend.rms_norm_dynamic_per_token_quant = dummyFusionOp(name="rms_norm_dynamic_per_token_quant")
|
|
|
|
|
|
__all__ = ["AscendQuickGELU", "AscendSiluAndMul", "AscendRotaryEmbedding", "AscendDeepseekScalingRotaryEmbedding"]
|