02886e2641
### What this PR does / why we need it?
This PR introduces support for W8A8 dynamic quantization for
Mixture-of-Experts (MoE) models on Ascend 310P devices. This is achieved
by:
- Implementing a new quantization scheme
`AscendW8A8DynamicFusedMoEMethod310`.
- Adding a unified MLP implementation (`unified_apply_mlp`) for 310P
that handles both quantized and unquantized paths.
- Refactoring the MoE and quantization configuration logic to correctly
route to the new 310P-specific implementations.
- Adding new e2e and unit tests to verify the functionality of MoE W8A8
quantization.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
- Added a new e2e test `test_qwen3_moe_tp2_w8a8` to test MoE W8A8
quantization in a multi-card setup.
- Added several new unit tests for the 310P-specific MoE components,
including `experts_selector`, `fused_moe`, `moe_comm_method`, `moe_mlp`,
and the new `w8a8_dynamic` quantization method.
- vLLM version: v0.15.0
- vLLM main:
d7e17aaacd
---------
Signed-off-by: pu-zhe <zpuaa@outlook.com>
67 lines
3.0 KiB
Python
67 lines
3.0 KiB
Python
#
|
|
# Copyright (c) 2026 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.
|
|
|
|
from unittest.mock import Mock, patch
|
|
|
|
import torch
|
|
|
|
from tests.ut.base import TestBase
|
|
from vllm_ascend._310p.quantization.methods.w8a8_dynamic import AscendW8A8DynamicFusedMoEMethod310
|
|
|
|
|
|
class TestAscendW8A8FusedMoEMethod310(TestBase):
|
|
num_experts = 8
|
|
hidden_size = 128
|
|
intermediate_size = 128
|
|
|
|
@patch("vllm_ascend._310p.quantization.methods.w8a8_dynamic.get_ep_group")
|
|
def setUp(self, mock_get_ep_group):
|
|
with patch(
|
|
"vllm_ascend._310p.quantization.methods.w8a8_dynamic.get_current_vllm_config"
|
|
) as mock_get_current_vllm_config:
|
|
mock_vllm_config = Mock()
|
|
mock_vllm_config.quant_config = Mock(quant_description={"group_size": 0})
|
|
mock_vllm_config.scheduler_config = Mock(
|
|
max_num_batched_tokens=2048, max_model_len=2048, enable_chunked_prefill=False
|
|
)
|
|
mock_get_current_vllm_config.return_value = mock_vllm_config
|
|
mock_ep_group = Mock()
|
|
mock_get_ep_group.return_value = mock_ep_group
|
|
mock_ascend_config = Mock()
|
|
|
|
mock_ascend_config.enable_chunked_prefill = False
|
|
|
|
self.quant_method = AscendW8A8DynamicFusedMoEMethod310()
|
|
|
|
def test_get_weight_310(self):
|
|
param_dict = self.quant_method.get_weight(
|
|
self.num_experts, self.intermediate_size, self.hidden_size, torch.float16
|
|
)
|
|
self.assertEqual(param_dict["w13_weight"].dtype, torch.int8)
|
|
self.assertEqual(
|
|
param_dict["w13_weight"].shape, (self.num_experts, 2 * self.intermediate_size, self.hidden_size)
|
|
)
|
|
self.assertEqual(param_dict["w2_weight"].dtype, torch.int8)
|
|
self.assertEqual(param_dict["w2_weight"].shape, (self.num_experts, self.hidden_size, self.intermediate_size))
|
|
|
|
def test_get_dynamic_quant_param_310(self):
|
|
param_dict = self.quant_method.get_dynamic_quant_param(
|
|
self.num_experts, self.intermediate_size, self.hidden_size, torch.float16
|
|
)
|
|
self.assertEqual(param_dict["w13_weight_scale"].dtype, torch.float32)
|
|
self.assertEqual(param_dict["w13_weight_scale"].shape, (self.num_experts, 2 * self.intermediate_size, 1))
|
|
self.assertEqual(param_dict["w2_weight_scale"].dtype, torch.float32)
|
|
self.assertEqual(param_dict["w2_weight_scale"].shape, (self.num_experts, self.hidden_size, 1))
|