Skip to content

[Bugfix][TPU] Fix CPU cache allocation #5869

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

Merged
merged 3 commits into from
Jun 26, 2024
Merged

[Bugfix][TPU] Fix CPU cache allocation #5869

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
16e9934
[Bugfix][TPU] Fix CPU cache allocation & swapping
WoosukKwon Jun 26, 2024
ed0c06c
Revert back to dynamo
WoosukKwon Jun 26, 2024
6ad7184
Revert back to dynamo
WoosukKwon Jun 26, 2024
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
5 changes: 2 additions & 3 deletions vllm/attention/backends/pallas.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -37,11 +37,10 @@ def swap_blocks(
) -> None:
src_k_cache, src_v_cache = src_kv_cache
dst_k_cache, dst_v_cache = dst_kv_cache
src_indices, dst_indices = src_to_dst
device = dst_k_cache.device
torch.ops.xla.dynamo_set_buffer_donor_(dst_k_cache, True)
torch.ops.xla.dynamo_set_buffer_donor_(dst_v_cache, True)

device = dst_k_cache.device
src_indices, dst_indices = src_to_dst
dst_k_cache[:, dst_indices] = src_k_cache[:, src_indices].to(device)
dst_v_cache[:, dst_indices] = src_v_cache[:, src_indices].to(device)

Expand Down
8 changes: 6 additions & 2 deletions vllm/worker/tpu_worker.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -156,14 +156,18 @@ def initialize_cache(
self.tpu_cache = []
tpu_cache_shape = self.model_runner.attn_backend.get_kv_cache_shape(
num_gpu_blocks, self.block_size, num_kv_heads, head_size)
cpu_cache_shape = self.model_runner.attn_backend.get_kv_cache_shape(
num_cpu_blocks, self.block_size, num_kv_heads, head_size)
for _ in range(num_layers):
tpu_k_cache = torch.zeros(tpu_cache_shape,
dtype=dtype,
device=self.device)
tpu_v_cache = torch.zeros_like(tpu_k_cache)
self.tpu_cache.append((tpu_k_cache, tpu_v_cache))
cpu_k_cache = torch.zeros_like(tpu_k_cache, device="cpu")
cpu_v_cache = torch.zeros_like(tpu_v_cache, device="cpu")
cpu_k_cache = torch.zeros(cpu_cache_shape,
dtype=dtype,
device="cpu")
cpu_v_cache = torch.zeros_like(cpu_k_cache)
self.cpu_cache.append((cpu_k_cache, cpu_v_cache))
self._warmup_model()

Expand Down