[MRV2] Miscellaneous code cleanup (#57980)
This commit is contained in:
@@ -11,7 +11,7 @@ from vllm.v1.worker.gpu.attn_utils import (
|
||||
build_slot_mappings_by_layer,
|
||||
)
|
||||
from vllm.v1.worker.gpu.block_table import BlockTables
|
||||
from vllm.v1.worker.gpu.cp_utils import maybe_prepare_dcp_local_seq_lens
|
||||
from vllm.v1.worker.gpu.cp_utils import prepare_dcp_local_seq_lens
|
||||
from vllm.v1.worker.gpu.cudagraph_utils import (
|
||||
AttentionState,
|
||||
BatchExecutionDescriptor,
|
||||
@@ -42,15 +42,16 @@ def _prepare_dflash_inputs_to_capture(
|
||||
attn_metadata = None
|
||||
if not skip_attn:
|
||||
query_start_loc_cpu = torch.from_numpy(input_batch.query_start_loc_np)
|
||||
input_batch.dcp_local_seq_lens = maybe_prepare_dcp_local_seq_lens(
|
||||
input_buffers.dcp_local_seq_lens,
|
||||
input_batch.seq_lens,
|
||||
input_batch.num_reqs,
|
||||
block_tables.cp_size,
|
||||
block_tables.cp_rank,
|
||||
block_tables.cp_interleave,
|
||||
num_reqs_padded=input_batch.num_reqs_after_padding,
|
||||
)
|
||||
if block_tables.cp_size > 1:
|
||||
input_batch.dcp_local_seq_lens = prepare_dcp_local_seq_lens(
|
||||
input_buffers.dcp_local_seq_lens,
|
||||
input_batch.seq_lens,
|
||||
input_batch.num_reqs,
|
||||
block_tables.cp_size,
|
||||
block_tables.cp_rank,
|
||||
block_tables.cp_interleave,
|
||||
num_reqs_padded=input_batch.num_reqs_after_padding,
|
||||
)
|
||||
attn_metadata = build_attn_metadata(
|
||||
attn_groups=attn_groups,
|
||||
num_reqs=num_reqs,
|
||||
|
||||
Reference in New Issue
Block a user