66b60c9440
### What this PR does / why we need it?
1. [Refact] Refact MLA/SFA weight prefetch to consist with moe weight
prefetch
2. Remove duplicated o_proj weight prefetch in forward for MLA/SFA
### Does this PR introduce _any_ user-facing change?
NA
### How was this patch tested?
1) Performance result:
Perf test data:
*) MLA:
| | 1st test | 2nd test | Output Token Throughput(Avg) | Performance
improvement percentage |
| --- | --- | --- | --- | --- |
| o_proj duplicate prefetch | 11.9669 token/s | 12.0287 token/s |
11.9978 |
| o_proj no duplicate prefetch | 12.5594 token/s | 12.6216 token/s |
12.5905 | 4.94%| |
single layer performace improve: 5%~8%
*) SFA:
| | 1st test | 2nd test | Output Token Throughput(Avg) | Performance
improvement percentage |
| --- | --- | --- | --- | --- |
| o_proj duplicate prefetch | 13.0523 token/s | 13.1084 token/s |
13.08035 | |
| o_proj no duplicate prefetch | 13.9844 token/s | 14.1678 token/s |
14.0761 | 7.6% |
- vLLM version: v0.15.0
- vLLM main:
d7e17aaacd
---------
Signed-off-by: leo-pony <nengjunma@outlook.com>
98 lines
3.0 KiB
Python
98 lines
3.0 KiB
Python
#
|
|
# 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.
|
|
#
|
|
|
|
from unittest.mock import MagicMock, patch
|
|
|
|
import pytest
|
|
import torch
|
|
from vllm.config import set_current_vllm_config
|
|
from vllm.model_executor.layers.activation import QuickGELU, SiluAndMul
|
|
|
|
from vllm_ascend.utils import AscendDeviceType
|
|
from vllm_ascend.utils import is_310p as is_310p_hw
|
|
|
|
|
|
@pytest.fixture
|
|
def dummy_tensor():
|
|
return torch.randn(4, 8, dtype=torch.float16)
|
|
|
|
|
|
@pytest.fixture
|
|
def default_vllm_config():
|
|
mock_config = MagicMock()
|
|
|
|
mock_config.compilation_config.dispatch_forward_backend = "eager"
|
|
|
|
mock_config.compilation_config.custom_ops = ["all"]
|
|
|
|
with set_current_vllm_config(mock_config):
|
|
yield mock_config
|
|
|
|
|
|
@patch("torch_npu.npu_fast_gelu", side_effect=lambda x: x + 1)
|
|
def test_QuickGELU_forward(mock_gelu, dummy_tensor, default_vllm_config):
|
|
layer = QuickGELU()
|
|
out = layer.forward(dummy_tensor)
|
|
|
|
expected_out = dummy_tensor + 1
|
|
assert torch.allclose(out, expected_out)
|
|
|
|
mock_gelu.assert_called_once()
|
|
|
|
|
|
@pytest.mark.skipif(is_310p_hw(), reason="non_310P device unittest case.")
|
|
@patch("vllm_ascend.ops.activation.get_weight_prefetch_method",
|
|
return_value=MagicMock())
|
|
@patch("torch_npu.npu_swiglu", side_effect=lambda x: x + 1)
|
|
def test_SiluAndMul_forward(
|
|
mock_swiglu,
|
|
mock_get_weight_prefetch_method,
|
|
dummy_tensor,
|
|
default_vllm_config,
|
|
):
|
|
layer = SiluAndMul()
|
|
out = layer.forward(dummy_tensor)
|
|
expected_arg = dummy_tensor
|
|
|
|
# assert mock_swiglu.call_count == 1
|
|
mock_swiglu.assert_called_once()
|
|
|
|
actual_arg = mock_swiglu.call_args[0][0]
|
|
assert torch.allclose(actual_arg, expected_arg), "npu_swiglu called with unexpected input"
|
|
|
|
expected_out = dummy_tensor + 1
|
|
assert torch.allclose(out, expected_out)
|
|
|
|
|
|
@pytest.mark.skipif(not is_310p_hw(), reason="310P device unittest case.")
|
|
@patch("torch.nn.functional.silu", side_effect=lambda x: x + 1)
|
|
def test_SiluAndMul_forward_310p(
|
|
mock_silu,
|
|
dummy_tensor,
|
|
default_vllm_config,
|
|
):
|
|
layer = SiluAndMul()
|
|
out = layer.forward(dummy_tensor)
|
|
h = dummy_tensor.shape[-1] // 2
|
|
expected_arg = dummy_tensor[..., :h]
|
|
|
|
# assert mock_silu.call_count == 1
|
|
mock_silu.assert_called_once()
|
|
|
|
actual_arg = mock_silu.call_args[0][0]
|
|
assert torch.allclose(actual_arg, expected_arg), "swiglu called with unexpected input"
|
|
|
|
expected_out = (dummy_tensor[..., :h] + 1) * dummy_tensor[..., h:]
|
|
assert torch.allclose(out, expected_out) |