fix(nsa): guard cp split to extend batches
This commit is contained in:
@@ -77,6 +77,31 @@ class TestPageAlignedCacheExtent(unittest.TestCase):
|
||||
|
||||
|
||||
class TestNSAInSeqCPUtils(unittest.TestCase):
|
||||
def test_can_cp_split_ignores_decode_batch_without_extend_lens(self):
|
||||
forward_batch = SimpleNamespace(
|
||||
forward_mode=ForwardMode.DECODE,
|
||||
extend_seq_lens_cpu=None,
|
||||
uses_cp_shared_kv=True,
|
||||
)
|
||||
|
||||
self.assertFalse(
|
||||
can_cp_split(
|
||||
seq_len=200, cp_size=8, use_nsa=True, forward_batch=forward_batch
|
||||
)
|
||||
)
|
||||
|
||||
def test_can_cp_split_fails_extend_batch_without_extend_lens(self):
|
||||
forward_batch = SimpleNamespace(
|
||||
forward_mode=ForwardMode.EXTEND,
|
||||
extend_seq_lens_cpu=None,
|
||||
uses_cp_shared_kv=True,
|
||||
)
|
||||
|
||||
with self.assertRaisesRegex(RuntimeError, "missing_extend_seq_lens_cpu"):
|
||||
can_cp_split(
|
||||
seq_len=200, cp_size=8, use_nsa=True, forward_batch=forward_batch
|
||||
)
|
||||
|
||||
def test_contiguous_valid_cp_query_count(self):
|
||||
from sglang.srt.layers.attention.nsa.nsa_indexer import (
|
||||
_compute_contiguous_valid_cp_query_count,
|
||||
|
||||
Reference in New Issue
Block a user