23bf5d4d48
### What this PR does / why we need it?
This pull request addresses a bug related to the fused mc2 functionality
within the EPLB (Expert Parallelism Load Balancing) system, specifically
impacting quantization and MoE communication.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.15.0
- vLLM main:
83b47f67b1
Signed-off-by: Spicy-Stick <873805887@qq.com>
Signed-off-by: root <root@localhost.localdomain>
42 lines
1.6 KiB
Python
42 lines
1.6 KiB
Python
import unittest
|
|
from unittest.mock import MagicMock, patch
|
|
|
|
import torch
|
|
|
|
from vllm_ascend.eplb.adaptor.vllm_adaptor import VllmEplbAdaptor
|
|
from vllm_ascend.quantization.methods.base import QuantType
|
|
from transformers import DeepseekV2Config
|
|
|
|
|
|
class TestVllmAdaptor(unittest.TestCase):
|
|
def setUp(self):
|
|
n_routed_experts = 256
|
|
mock_model = MagicMock()
|
|
mock_model.model.named_parameters.return_value = dict()
|
|
config = DeepseekV2Config(n_routed_experts=n_routed_experts)
|
|
mock_model.config = config
|
|
mock_model.get_expert_map.return_value = [i for i in range(n_routed_experts)]
|
|
mock_model.get_log2phy_map.return_value = [i for i in range(n_routed_experts)]
|
|
self.model = mock_model
|
|
num_dense_layers = getattr(config, "first_k_dense_replace", 0)
|
|
self.model.model.layers[num_dense_layers].mlp.experts.quant_type = QuantType.W8A8
|
|
|
|
self.mock_rank = patch("vllm_ascend.eplb.adaptor.vllm_adaptor.dist.get_rank", return_value=0).start()
|
|
self.mock_size = patch("vllm_ascend.eplb.adaptor.vllm_adaptor.dist.get_world_size", return_value=4).start()
|
|
|
|
@patch("torch.empty_like", return_value=torch.zeros(16, 32))
|
|
def test_init_fp16(self, mock_func):
|
|
self.model.quant_config = None
|
|
VllmEplbAdaptor(self.model)
|
|
|
|
@patch("torch.empty_like", return_value=torch.zeros(16, 32))
|
|
def test_init_w8a8(self, mock_func):
|
|
VllmEplbAdaptor(self.model)
|
|
|
|
def tearDown(self):
|
|
self.mock_rank.stop()
|
|
self.mock_size.stop()
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|
|
|