Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[V1][TPU] Remove unnecessary padding for running on TPU. #14467

Merged
merged 3 commits into from
Mar 9, 2025
+6 −18
Merged

[V1][TPU] Remove unnecessary padding for running on TPU. #14467

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
ad38ffa
Reduce the size of block_table by getting rid of padding.
vanbasten23 Mar 8, 2025
c881174
remove unused import
vanbasten23 Mar 8, 2025
541afd6
remove other padding
vanbasten23 Mar 8, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions vllm/v1/attention/backends/pallas.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,8 +12,8 @@
from vllm.attention.backends.utils import CommonAttentionState

# These are the 2 tunable parameters of the paged attention Pallas kernel.
NUM_QUERIES_PER_BLOCK = 16
NUM_KV_PAGES_PER_BLOCK = 256
NUM_QUERIES_PER_BLOCK = 32
NUM_KV_PAGES_PER_BLOCK = 128


class PallasAttentionBackend(AttentionBackend):
Expand Down
10 changes: 1 addition & 9 deletions vllm/v1/worker/tpu_model_runner.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -141,16 +141,8 @@ def __init__(
device="cpu")
self.slot_mapping_np = self.slot_mapping_cpu.numpy()

# self.input_batch.block_table has a shape of [max_num_reqs,
# max_num_blocks_per_req]. To reduce the number of recompilation,
# we want the block_table.shape[0] to be num_tokens.
# To make the block_table to be compatible with the paged attention
# kernel, we want the block_table[1] to be multiple of
# NUM_KV_PAGES_PER_BLOCK.
padded_max_num_blocks_per_req = _get_padded_number(
self.max_num_blocks_per_req, NUM_KV_PAGES_PER_BLOCK)
self.block_table_cpu = torch.zeros(
(self.max_num_tokens, padded_max_num_blocks_per_req),
(self.max_num_tokens, self.max_num_blocks_per_req),
dtype=self.input_batch.block_table.get_cpu_tensor().dtype,
device="cpu")

Expand Down
Loading