biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Attention][1/n] Remove usage of deprecated seq_lens_cpu and num_computed_tokens_cpu CommonAttentionMetadata properties (#31773)

Browse Source 
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
This commit is contained in:
Lucas Wilkinson
2026-01-06 07:05:17 -05:00
committed by GitHub
parent 14df02b4e1
commit e0327c9db2
9 changed files with 23 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
tests/v1/attention/test_attention_backends.py
View File

@@ -126,12 +126,12 @@ def create_and_prepopulate_kv_cache(
Tuple of (kv_cache, updated_block_table)
"""
batch_size = len(k_contexts)
seq_lens = common_attn_metadata.seq_lens_cpu
seq_lens = common_attn_metadata.seq_lens.cpu()
query_lens = (
common_attn_metadata.query_start_loc_cpu[1:]
- common_attn_metadata.query_start_loc_cpu[:-1]
)
context_lens = common_attn_metadata.num_computed_tokens_cpu
context_lens = seq_lens - query_lens
block_table = common_attn_metadata.block_table_tensor
slot_mapping = common_attn_metadata.slot_mapping

4
tests/v1/attention/test_mla_backends.py
View File

@@ -154,12 +154,12 @@ def create_and_prepopulate_kv_cache(
MLA KV cache tensor
"""
batch_size = len(kv_c_contexts)
seq_lens = common_attn_metadata.seq_lens_cpu
seq_lens = common_attn_metadata.seq_lens.cpu()
query_lens = (
common_attn_metadata.query_start_loc_cpu[1:]
- common_attn_metadata.query_start_loc_cpu[:-1]
)
context_lens = common_attn_metadata.num_computed_tokens_cpu
context_lens = seq_lens - query_lens
block_table = common_attn_metadata.block_table_tensor
slot_mapping = common_attn_metadata.slot_mapping

2
tests/v1/attention/test_sparse_mla_backends.py
View File

@@ -297,7 +297,7 @@ def test_sparse_backend_decode_correctness(
positions = np.arange(starts[-1], dtype=np.int32) - np.repeat(
starts[:-1], seg_lengths
)
seq_lengths = np.asarray(common_attn_metadata.seq_lens_cpu, dtype=np.int32)
seq_lengths = np.asarray(common_attn_metadata.seq_lens.cpu(), dtype=np.int32)
prefix_lengths = seq_lengths - seg_lengths
positions += np.repeat(prefix_lengths, seg_lengths)