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

Upgrade CANN to 8.3.rc1 (#3945)

Browse Source 
### What this PR does / why we need it?
This PR upgrade CANN from 8.2rc1 to 8.3rc1 and remove the CANN version
check logic.

TODO: we notice that UT runs failed with CANN 8.3 image. So the base
image for UT is still 8.2. We'll fix it later.


- vLLM version: v0.11.0
- vLLM main:
83f478bb19

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
wangxiyuan
2025-11-03 20:21:07 +08:00
committed by GitHub
parent 49d74785c4
commit cc2cd42ad3
39 changed files with 119 additions and 213 deletions
Download Patch File Download Diff File Expand all files Collapse all files

47
tests/ut/attention/test_attention_mask.py
View File

@@ -74,11 +74,10 @@ class TestAttentionMaskBuilder(TestBase):
attn_mask = attention_mask_builder.get_attn_mask(
max_seq_len=2048, dtype=torch.float16, device=torch.device("cpu"))
self.assertEqual(attn_mask.shape, (2048, 2048))
self.assertEqual(attn_mask[0][-1],
torch.tensor(float("-inf"), dtype=torch.float16))
self.assertEqual(attention_mask_builder._seq_len_cached, 2048)
self.assertEqual(attn_mask[0][-1], torch.tensor(True))
self.assertEqual(attention_mask_builder._seq_len_cached, 1024)
self.assertEqual(attention_mask_builder.attn_mask_cache.shape,
(2048, 2048))
(1024, 1024))
self.assertEqual(attention_mask_builder.attn_mask_cache[0][-1],
torch.tensor(float("-inf"), dtype=torch.float16))
@@ -91,43 +90,5 @@ class TestAttentionMaskBuilder(TestBase):
dtype=torch.float16,
device=torch.device("cpu"),
)
self.assertEqual(attn_mask.shape, (6, 100))
self.assertEqual(attn_mask.shape, (2048, 2048))
self.assertEqual(attention_mask_builder._seq_len_cached, 1024)
attn_mask = attention_mask_builder.get_splitfuse_attn_mask(
seq_lens=torch.tensor([10, 3000, 2000]),
position=torch.tensor([7, 8, 9, 2999, 1999]),
dtype=torch.float16,
device=torch.device("cpu"),
)
self.assertEqual(attn_mask.shape, (5, 3000))
self.assertEqual(attention_mask_builder._seq_len_cached, 3000)
# splitfuse_attn_mask now only supports data types: torch.float16 and torch.bfloat16
# otherwise raise ValueError
with self.assertRaises(ValueError):
attn_mask = attention_mask_builder.get_splitfuse_attn_mask(
seq_lens=torch.tensor([10, 20, 100]),
position=torch.tensor([7, 8, 9, 18, 19, 99]),
dtype=torch.int8,
device=torch.device("cpu"),
)
def test_mask_value_cleanliness(self):
attention_mask_builder = AttentionMaskBuilder(max_seq_len=6,
dtype=torch.bfloat16)
self.assertEqual(attention_mask_builder.attn_mask_cache[-2][-1],
torch.tensor(1, dtype=torch.bfloat16))
attn_mask = attention_mask_builder.get_splitfuse_attn_mask(
seq_lens=torch.tensor([6]),
position=torch.tensor([3, 4, 5]),
dtype=torch.bfloat16,
device=torch.device("cpu"),
)
self.assertEqual(
attn_mask[-2][-1],
torch.tensor(-10000, dtype=torch.bfloat16,
device=attn_mask.device))
self.assertEqual(attention_mask_builder.attn_mask_cache[-2][-1],
torch.tensor(1, dtype=torch.bfloat16))