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

[Misc] Remove CP Redundant Variables after FIA operator enables for CANN 8.5 (#6013)

Browse Source 
### What this PR does / why we need it?
PCP/DCP splits the kv-cache onto different cards. After introducing the
parameter cp-kv-cache-interleave-size, the first size tokens will be
cached at Card 0, and so on.
However, if there are too few tokens, some cards will not store the
key-value pairs, resulting in values ​​of 0, corrupted values, and
precision issues. Currently, additional operations are introduced to
avoid this precision problem.

After we integrate FIA operator in mla_cp._forward_decode and CANN
updates to 8.5.0, we now can remove these additional operations.
### Does this PR introduce _any_ user-facing change?

### How was this patch tested?
passed all CI by CANN 8.5.0
- vLLM version: v0.13.0
- vLLM main:
2c24bc6996

Signed-off-by: dsxsteven <dsxsteven@sina.com>
Signed-off-by: dsxsteven <36877507+dsxsteven@users.noreply.github.com>
This commit is contained in:
dsxsteven
2026-01-23 14:13:12 +08:00
committed by GitHub
parent 418a43e2a2
commit 8378bc28b0
8 changed files with 78 additions and 57 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
tests/ut/attention/test_mla_cp.py
View File

@@ -439,11 +439,7 @@ class TestAscendMLAImpl(TestBase):
decode_metadata = MagicMock()
decode_metadata.actual_seq_lengths_q = MagicMock()
decode_metadata.seq_lens_list = MagicMock()
decode_metadata.batch_seq_mask = torch.tensor([True, False],
dtype=torch.bool)
result = _process_attn_out_lse(attn_output, softmax_lse,
decode_metadata.batch_seq_mask)
result = _process_attn_out_lse(attn_output, softmax_lse)
self.assertEqual(result.shape[0], B * self.impl.pcp_size)
self.assertEqual(result.shape[1], N)
@@ -478,8 +474,6 @@ class TestAscendMLAImpl(TestBase):
attn_metadata.decode = MagicMock()
attn_metadata.decode.actual_seq_lengths_q = MagicMock()
attn_metadata.decode.seq_lens_list = MagicMock()
attn_metadata.decode.batch_seq_mask = torch.tensor([False, False],
dtype=torch.bool)
self.impl.enable_kv_nz = True
@@ -886,12 +880,9 @@ class TestAscendMLAImpl(TestBase):
# Inputs
attn_output = torch.randn(B, H, D)
softmax_lse = torch.randn(B, H, 1)
batch_seq_mask = torch.tensor([False, True, False, False]) # [B]
decode_meta = MagicMock()
decode_meta.batch_seq_mask = batch_seq_mask
result = _process_attn_out_lse(attn_output, softmax_lse,
batch_seq_mask)
result = _process_attn_out_lse(attn_output, softmax_lse)
# [PCP * S, DCP * H, D + 1]
self.assertIsInstance(result, torch.Tensor)
assert result.shape == (B * self.impl.pcp_size, H, D + 1)

5
tests/ut/attention/test_mla_v1.py
View File

@@ -137,7 +137,6 @@ class TestAscendMLADecodeMetadata(TestBase):
seq_lens_list = [2, 3]
attn_mask = None
cp_seq_len = torch.tensor([2, 3])
batch_seq_mask = torch.tensor([[1, 1, 0, 0], [1, 1, 1, 0]])
metadata = AscendMLADecodeMetadata(input_positions=input_positions,
block_table=block_table,
@@ -145,8 +144,7 @@ class TestAscendMLADecodeMetadata(TestBase):
max_seq_lens=max_seq_lens,
seq_lens_list=seq_lens_list,
attn_mask=attn_mask,
cp_seq_len=cp_seq_len,
batch_seq_mask=batch_seq_mask)
cp_seq_len=cp_seq_len)
self.assertIs(metadata.input_positions, input_positions)
self.assertIs(metadata.block_table, block_table)
@@ -155,7 +153,6 @@ class TestAscendMLADecodeMetadata(TestBase):
self.assertEqual(metadata.seq_lens_list, seq_lens_list)
self.assertIsNone(attn_mask)
self.assertIs(metadata.cp_seq_len, cp_seq_len)
self.assertIs(metadata.batch_seq_mask, batch_seq_mask)
class TestAscendMLAMetadata(TestBase):