[Deprecation] Deprecate items scheduled for 0.29 (#55353)

Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
Wentao Ye
2026-09-11 08:17:02 -07:00
committed by GitHub
parent 1e1060f998
commit 5fe77aecfc
28 changed files with 62 additions and 305 deletions
@@ -115,7 +115,7 @@ def create_cross_attention_backend(
# needed here to know how many tokens to attend to from the cached
# cross-attention KV cache.
new_metadata.seq_lens = common_attn_metadata.encoder_seq_lens
new_metadata._seq_lens_cpu = torch.from_numpy(
new_metadata.seq_lens_cpu_upper_bound = torch.from_numpy(
common_attn_metadata.encoder_seq_lens_cpu
)
+2 -4
View File
@@ -161,10 +161,8 @@ def create_whisper_attention_backend_with_block_pooling(
new_common_attn_metadata.query_start_loc *= block_pool_size
new_common_attn_metadata.query_start_loc_cpu *= block_pool_size
new_common_attn_metadata.seq_lens *= block_pool_size
if new_common_attn_metadata._seq_lens_cpu is not None:
new_common_attn_metadata._seq_lens_cpu *= block_pool_size
if new_common_attn_metadata._num_computed_tokens_cpu is not None:
new_common_attn_metadata._num_computed_tokens_cpu *= block_pool_size
if new_common_attn_metadata.seq_lens_cpu_upper_bound is not None:
new_common_attn_metadata.seq_lens_cpu_upper_bound *= block_pool_size
new_common_attn_metadata.num_actual_tokens *= block_pool_size
new_common_attn_metadata.max_query_len *= block_pool_size
new_common_attn_metadata.max_seq_len *= block_pool_size