minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[Fix] Unify ForwardBatch extend lens cpu fields to their declared list type (#30896)

Browse Source
This commit is contained in:
Liangsheng Yin
2026-07-11 17:30:53 -05:00
committed by GitHub
parent d8ef76682e
commit 4884f6fbee
6 changed files with 12 additions and 29 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/manual/attention/test_flashattn_backend.py
+2 -6
View File
@@ -222,15 +222,11 @@ class TestFlashAttentionBackend(CustomTestCase):
extend_prefix_lens=torch.tensor(
[prefix_len] * self.batch_size, device=self.device
),
extend_prefix_lens_cpu=torch.tensor(
[prefix_len] * self.batch_size, device="cpu"
),
extend_prefix_lens_cpu=[prefix_len] * self.batch_size,
extend_seq_lens=torch.tensor(
[q_len] * self.batch_size, device=self.device
),
extend_seq_lens_cpu=torch.tensor(
[q_len] * self.batch_size, device="cpu"
),
extend_seq_lens_cpu=[q_len] * self.batch_size,
)
if attn_cp_size > 1:
forward_batch.attn_cp_metadata = type(
test/manual/attention/test_flashattn_mla_backend.py
+2 -6
View File
@@ -190,15 +190,11 @@ class TestFlashAttentionMLABackend(CustomTestCase):
extend_prefix_lens=torch.tensor(
[prefix_len] * self.batch_size, device=self.device
),
extend_prefix_lens_cpu=torch.tensor(
[prefix_len] * self.batch_size, device="cpu"
),
extend_prefix_lens_cpu=[prefix_len] * self.batch_size,
extend_seq_lens=torch.tensor(
[q_len] * self.batch_size, device=self.device
),
extend_seq_lens_cpu=torch.tensor(
[q_len] * self.batch_size, device="cpu"
),
extend_seq_lens_cpu=[q_len] * self.batch_size,
)
else: # ForwardMode.DECODE
test/registered/kernels/test_dsa_indexer.py
+2 -4
View File
@@ -367,11 +367,9 @@ class TestDSAIndexer(CustomTestCase):
extend_prefix_lens=torch.tensor(
[total_len - q_len] * batch_size, device=self.device
),
extend_prefix_lens_cpu=torch.tensor(
[total_len - q_len] * batch_size, device="cpu"
),
extend_prefix_lens_cpu=[total_len - q_len] * batch_size,
extend_seq_lens=torch.tensor([q_len] * batch_size, device=self.device),
extend_seq_lens_cpu=torch.tensor([q_len] * batch_size, device="cpu"),
extend_seq_lens_cpu=[q_len] * batch_size,
)
else: # ForwardMode.DECODE
decode_len = 1