Skip to content

fix(worker): optimize swap_states to copy only active token prefixes #34733

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
njhill merged 6 commits into from
Mar 18, 2026
+20 −8
Merged

fix(worker): optimize swap_states to copy only active token prefixes #34733

Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
3caf9b1
fix: optimize swap_states to copy only active token prefixes
phiott Feb 17, 2026
2e69161
Merge branch 'main' into swap-states-active-prefix
phiott Feb 21, 2026
d67fb65
Merge branch 'main' into swap-states-active-prefix
phiott Mar 2, 2026
b85a6db
Merge branch 'main' into swap-states-active-prefix
LucasWilkinson Mar 3, 2026
307e460
Merge branch 'main' into swap-states-active-prefix
phiott Mar 13, 2026
01fcaf7
Merge branch 'main' into swap-states-active-prefix
phiott Mar 18, 2026
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
28 changes: 20 additions & 8 deletions vllm/v1/worker/gpu_input_batch.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -529,6 +529,12 @@ def remove_request(self, req_id: str) -> int | None:
def swap_states(self, i1: int, i2: int) -> None:
old_id_i1 = self._req_ids[i1]
old_id_i2 = self._req_ids[i2]
# Only swap the active token prefix for each request. Copying full
# max_model_len rows is expensive and unnecessary during reordering.
i1_active_token_count = self._get_active_token_count(i1)
i2_active_token_count = self._get_active_token_count(i2)
max_active_token_count = max(i1_active_token_count, i2_active_token_count)

self._req_ids[i1], self._req_ids[i2] = self._req_ids[i2], self._req_ids[i1] # noqa
self.req_output_token_ids[i1], self.req_output_token_ids[i2] = (
self.req_output_token_ids[i2],
Expand Down Expand Up @@ -560,12 +566,15 @@ def swap_states(self, i1: int, i2: int) -> None:
# self.token_ids_cpu[i1, ...], self.token_ids_cpu[i2, ...], =\
# self.token_ids_cpu[i2, ...], self.token_ids_cpu[i1, ...]
# instead, we need to temporarily copy the data for one of the indices
# TODO(lucas): optimize this by only copying valid indices
tmp = self.token_ids_cpu[i1, ...].copy()
self.token_ids_cpu[i1, ...] = self.token_ids_cpu[i2, ...]
self.token_ids_cpu[i2, ...] = tmp
tmp_token_ids = self.token_ids_cpu[i1, :max_active_token_count].copy()
self.token_ids_cpu[i1, :max_active_token_count] = self.token_ids_cpu[
i2, :max_active_token_count
]
self.token_ids_cpu[i2, :max_active_token_count] = tmp_token_ids

self.is_token_ids[[i1, i2], ...] = self.is_token_ids[[i2, i1], ...]
self.is_token_ids[[i1, i2], :max_active_token_count] = self.is_token_ids[
[i2, i1], :max_active_token_count
]

# Swap prompt embeddings if they exist
embeds_i1 = self.req_prompt_embeds.get(i1)
Expand Down Expand Up @@ -629,6 +638,11 @@ def swap_states(self, i1: int, i2: int) -> None:
self.allowed_token_ids_mask_cpu_tensor[i1],
)

def _get_active_token_count(self, req_index: int) -> int:
return int(self.num_tokens_no_spec[req_index]) + len(
self.spec_token_ids[req_index]
)

def condense(self) -> None:
"""Slide non-empty requests down into lower, empty indices.

Expand Down Expand Up @@ -678,9 +692,7 @@ def condense(self) -> None:
self.req_output_token_ids[last_req_index] = None
self.req_id_to_index[req_id] = empty_index

num_tokens = self.num_tokens_no_spec[last_req_index] + len(
self.spec_token_ids[last_req_index]
)
num_tokens = self._get_active_token_count(last_req_index)

(self.spec_token_ids[last_req_index], self.spec_token_ids[empty_index]) = (
self.spec_token_ids[empty_index],
Expand Down
Loading