Skip to content

Improving Total Token Throughput by 1%: Reducing CPU Overhead in Zero-Overhead Scheduling #4790

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

Closed
wants to merge 5 commits into from
Closed

Improving Total Token Throughput by 1%: Reducing CPU Overhead in Zero-Overhead Scheduling #4790

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
05d057f
delete_launch_done
WANG-GH Mar 26, 2025
ad95bd2
Merge branch 'main' into delete_launch_done
WANG-GH Mar 26, 2025
3b58c29
Merge branch 'main' into delete_launch_done
zhaochenyang20 Mar 27, 2025
fdb2a33
Merge branch 'main' into delete_launch_done
merrymercy Mar 28, 2025
d4ca039
Merge branch 'main' into delete_launch_done
xiezhq-hermann Mar 30, 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
2 changes: 0 additions & 2 deletions python/sglang/srt/managers/tp_worker.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -173,8 +173,6 @@ def forward_batch_generation(
) -> Tuple[LogitsProcessorOutput, Optional[torch.Tensor]]:
forward_batch = ForwardBatch.init_new(model_worker_batch, self.model_runner)
logits_output = self.model_runner.forward(forward_batch)
if launch_done:
launch_done.set()

if skip_sample:
next_token_ids = None
Expand Down
4 changes: 1 addition & 3 deletions python/sglang/srt/managers/tp_worker_overlap_thread.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -132,7 +132,6 @@ def forward_thread_func_(self):
batch_pt += 1

# Create event
self.launch_done = threading.Event()
copy_done = torch.get_device_module(self.device).Event()

# Resolve future tokens in the input
Expand All @@ -141,7 +140,7 @@ def forward_thread_func_(self):

# Run forward
logits_output, next_token_ids = self.worker.forward_batch_generation(
model_worker_batch, self.launch_done
model_worker_batch
)

# Update the future token ids map
Expand Down Expand Up @@ -171,7 +170,6 @@ def forward_thread_func_(self):
def resolve_batch_result(self, bid: int):
copy_done, logits_output, next_token_ids = self.output_queue.get()
copy_done.synchronize()
self.launch_done.wait()

if logits_output.next_token_logprobs is not None:
logits_output.next_token_logprobs = (
Expand Down