f7b904641e
### What this PR does / why we need it?
Upgrade vllm commit to 0109 (bde38c11df0ea066a740efe9b77fff5418be45df)
1. remove `init_cached_hf_modules ` due to
https://github.com/vllm-project/vllm/pull/31786
2. fix spec_decode e2e test due to
https://github.com/vllm-project/vllm/pull/29821 break
3. fix `vllm.v1.attention.backends.utils` duo to
https://github.com/vllm-project/vllm/pull/31891
4. fix `self.seq_lens - query_lens` on same device due to
https://github.com/vllm-project/vllm/pull/31773
5. skip model_runner_v2 e2e test due to `'_OpNamespace' '_C' object has
no attribute 'get_cuda_view_from_cpu_tensor'`
- vLLM version: v0.13.0
- vLLM main:
2f4e6548ef
Signed-off-by: hfadzxy <starmoon_zhang@163.com>
68 lines
2.3 KiB
Python
68 lines
2.3 KiB
Python
from unittest.mock import MagicMock, patch
|
|
|
|
import pytest
|
|
import torch
|
|
from vllm.config import set_current_vllm_config
|
|
from vllm.model_executor.layers.layernorm import RMSNorm
|
|
|
|
from vllm_ascend.utils import AscendDeviceType
|
|
|
|
|
|
@pytest.fixture
|
|
def dummy_tensor():
|
|
return torch.randn(4, 8, dtype=torch.float16)
|
|
|
|
|
|
def mock_rms_norm(x, weight, eps):
|
|
return x + 1, None
|
|
|
|
|
|
def mock_add_rms_norm(x, residual, weight, eps):
|
|
return 2 * x, None, 2 * residual
|
|
|
|
|
|
@pytest.fixture(autouse=True)
|
|
def default_vllm_config():
|
|
mock_config = MagicMock()
|
|
mock_config.compilation_config.custom_ops = ["all"]
|
|
|
|
with set_current_vllm_config(mock_config):
|
|
yield mock_config
|
|
|
|
|
|
@pytest.mark.parametrize("is_310p", [True, False])
|
|
@pytest.mark.parametrize("residual",
|
|
[None, torch.randn(4, 8, dtype=torch.float32)])
|
|
@patch("torch_npu.npu_rms_norm", side_effect=mock_rms_norm)
|
|
@patch("torch_npu.npu_add_rms_norm", side_effect=mock_add_rms_norm)
|
|
def test_RMSNorm_forward(mock_add_rmsnorm, mock_rmsnorm, is_310p, residual,
|
|
dummy_tensor, default_vllm_config):
|
|
|
|
with patch("vllm_ascend.utils.get_ascend_device_type",
|
|
return_value=AscendDeviceType._310P
|
|
if is_310p else AscendDeviceType.A3):
|
|
layer = RMSNorm(hidden_size=8, eps=1e-05)
|
|
if residual is not None:
|
|
out_x, out_residual = layer.forward_oot(dummy_tensor, residual)
|
|
|
|
if is_310p:
|
|
expected_arg_x = dummy_tensor + residual.to(dummy_tensor.dtype)
|
|
expected_out_x = expected_arg_x + 1
|
|
expected_out_residual = expected_arg_x.to(residual.dtype)
|
|
|
|
mock_rmsnorm.assert_called_once()
|
|
assert torch.allclose(out_x, expected_out_x)
|
|
assert torch.allclose(out_residual, expected_out_residual)
|
|
else:
|
|
expected_out_x = 2 * dummy_tensor
|
|
expected_out_residual = 2 * residual
|
|
mock_add_rmsnorm.assert_called_once()
|
|
assert torch.allclose(out_x, expected_out_x)
|
|
assert torch.allclose(out_residual, expected_out_residual)
|
|
else:
|
|
out_x = layer.forward_oot(dummy_tensor, residual)
|
|
expected_out_x = dummy_tensor + 1
|
|
|
|
mock_rmsnorm.assert_called_once()
|
|
assert torch.allclose(out_x, expected_out_x)
|