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] Increase default batch size for H100/H200 #12369

Merged
merged 2 commits into from
Jan 24, 2025
Merged

[V1] Increase default batch size for H100/H200 #12369

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
abf9093
[V1] Increase default batch size for H100/H200
WoosukKwon Jan 23, 2025
26f1d3d
Increase LLM batch size
WoosukKwon Jan 23, 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
21 changes: 16 additions & 5 deletions vllm/engine/arg_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1279,11 +1279,22 @@ def _override_v1_engine_args(self, usage_context: UsageContext) -> None:
self.enable_chunked_prefill = True
# When no user override, set the default values based on the usage
# context.
# TODO(woosuk): Tune the default values for different hardware.
default_max_num_batched_tokens = {
UsageContext.LLM_CLASS: 8192,
UsageContext.OPENAI_API_SERVER: 2048,
}
# Use different default values for different hardware.
from vllm.platforms import current_platform
device_name = current_platform.get_device_name().lower()
if "h100" in device_name or "h200" in device_name:
# For H100 and H200, we use larger default values.
default_max_num_batched_tokens = {
UsageContext.LLM_CLASS: 16384,
UsageContext.OPENAI_API_SERVER: 8192,
}
else:
# TODO(woosuk): Tune the default values for other hardware.
default_max_num_batched_tokens = {
UsageContext.LLM_CLASS: 8192,
UsageContext.OPENAI_API_SERVER: 2048,
}
Comment on lines +1292 to +1296
Copy link
Member

@ywang96 ywang96 Jan 24, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Whenever we tune this/work on this TODO, let's also make a method out of it instead of putting the logic inside arg_utils.py

Something like

default_max_num_batched_tokens = auto_tuned_batch_size(UsageContext, current_platform)


if (self.max_num_batched_tokens is None
and usage_context in default_max_num_batched_tokens):
self.max_num_batched_tokens = default_max_num_batched_tokens[
Expand Down
Loading