A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 02:15 UTC
[TPU][Bugfix] fix test_pallas (#20666)
Signed-off-by: Chengji Yao <chengjiyao@google.com>
Chengji Yao committed
Jul 9, 2025 at 09:32 UTC
eb58f5953de8af1b02e4319519d6b9b806b1bbff
1 file changed
+2
tests/v1/tpu/test_pallas.py
+2
@@ -50,6 +50,7 @@ def test_ragged_paged_attention():
50
slot_mapping = torch.zeros((3, num_tokens), dtype=torch.int64)
51
max_num_reqs = 8
52
max_num_blocks_per_req = 8
53
+ num_kv_update_slices = torch.tensor([num_tokens], dtype=torch.int32)
54
block_tables = torch.zeros((max_num_reqs, max_num_blocks_per_req),
55
dtype=torch.int32)
56
context_lens = torch.ones((max_num_reqs, ), dtype=torch.int32)
@@ -65,6 +66,7 @@ def test_ragged_paged_attention():
66
context_lens=context_lens,
67
query_start_loc=query_start_loc,
68
num_seqs=num_seqs,
69
+ num_kv_update_slices=num_kv_update_slices,
70
num_slices_per_kv_cache_update_block=8,
71
)
72