EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[Bugfix] bugfix for moe_mlp in vllm-ascend/v0.11.0-dev (#4885)

Browse Source 
### What this PR does / why we need it?
This PR fixes a bug in the moe_mlp module by correcting the arguments
passed to the torch_npu.npu_dequant_swiglu_quant function.It properly
converts group_list from a cumulative sum to counts for the group_index
parameter.

### Does this PR introduce _any_ user-facing change?
No


- vLLM version: v0.12.0
- vLLM main: https://github.com/vllm-project/vllm/main

---------

Signed-off-by: tanqingshan (A)  <50050625@china.huawei.com>
Signed-off-by: tanqingshan (A) <50050625@china.huawei.com>
Co-authored-by: tanqingshan (A) <50050625@china.huawei.com>
Co-authored-by: Mercykid-bash <ruanche0218@gmail.com>
This commit is contained in:
Clorist33
2025-12-12 14:51:47 +08:00
committed by GitHub
parent 9c0ad46c1a
commit 4f0dddc9ee
5 changed files with 41 additions and 34 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
tests/ut/ops/test_expert_load_balancer.py
View File

@@ -82,7 +82,7 @@ class TestExpertLoadBalancer(TestBase):
)
self.assertEqual(expert_placement_map.shape,
(self.expert_load_balancer.layers_num,
self.expert_load_balancer.ranks_num, 10))
self.expert_load_balancer.ranks_num, 8))
self.assertTrue(torch.all(expert_placement_map >= -1))
def test_generate_log2phy_expert_map(self):
@@ -90,7 +90,7 @@ class TestExpertLoadBalancer(TestBase):
log2phy_map = self.expert_load_balancer.generate_log2phy_expert_map(
layer_id)
self.assertEqual(log2phy_map.shape,
(self.expert_load_balancer.ranks_num, 10))
(self.expert_load_balancer.ranks_num, 8))
self.assertTrue(torch.all(log2phy_map >= -1))
@mock.patch("torch_npu.npu._lazy_init")
@@ -101,7 +101,7 @@ class TestExpertLoadBalancer(TestBase):
rank_local_expert_num, rank_expert_map = self.expert_load_balancer.get_rank_placement_map(
layer_id, rank_id)
self.assertEqual(rank_local_expert_num, 5)
expected_tensor = torch.tensor([2, -1, 1, 3, -1, 4, -1, 0, -1, -1],
expected_tensor = torch.tensor([2, -1, 1, 3, -1, 4, -1, 0],
dtype=torch.int32).to(
rank_expert_map.device)
self.assertTrue(rank_expert_map.equal(expected_tensor))
@@ -109,7 +109,7 @@ class TestExpertLoadBalancer(TestBase):
rank_id = 1
rank_local_expert_num, rank_expert_map = self.expert_load_balancer.get_rank_placement_map(
layer_id, rank_id)
expected_tensor = torch.tensor([-1, 1, 4, -1, 2, -1, 0, 3, -1, -1],
expected_tensor = torch.tensor([-1, 1, 4, -1, 2, -1, 0, 3],
dtype=torch.int32).to(
rank_expert_map.device)
self.assertTrue(rank_expert_map.equal(expected_tensor))
@@ -119,7 +119,7 @@ class TestExpertLoadBalancer(TestBase):
rank_id = 0
log2phy_map = self.expert_load_balancer.get_rank_log2phy_map(
layer_id, rank_id)
expected_tensor = torch.tensor([2, 6, 1, 3, 7, 4, 5, 0, -1, -1],
expected_tensor = torch.tensor([2, 6, 1, 3, 7, 4, 5, 0],
dtype=torch.int32).to(
log2phy_map.device)
self.assertTrue(log2phy_map.equal(expected_tensor))
@@ -127,7 +127,7 @@ class TestExpertLoadBalancer(TestBase):
rank_id = 1
log2phy_map = self.expert_load_balancer.get_rank_log2phy_map(
layer_id, rank_id)
expected_tensor = torch.tensor([2, 6, 9, 3, 7, 4, 5, 8, -1, -1],
expected_tensor = torch.tensor([2, 6, 9, 3, 7, 4, 5, 8],
dtype=torch.int32).to(
log2phy_map.device)
self.assertTrue(log2phy_map.equal(expected_tensor))