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

Jump to bottom

[BugFix][Habana_main][Multistep]Fix multistep deepcopy overhead #452

Open
wants to merge 2 commits into
base: habana_main
Choose a base branch
from
+16 −2
Open

[BugFix][Habana_main][Multistep]Fix multistep deepcopy overhead #452

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
c948323
Fix multistep deepcopy overhead
xuechendi Nov 1, 2024
50515b0
Fix formatting issue detected by yapf, ruff, isort and mypy
xuechendi Nov 1, 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
18 changes: 16 additions & 2 deletions vllm/worker/hpu_model_runner.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2109,6 +2109,7 @@ def execute_model(
# we only want to pythonize in the last step
sampling_metadata.skip_sampler_cpu_output = True
self.model.model.sampler.include_gpu_probs_tensor = True
cache_orig_output_token_ids: List[Dict] = []
for i in range(num_steps):
with self.profiler.record_event('internal', model_event_name):
hidden_states = self.model.forward(
Expand Down Expand Up @@ -2159,8 +2160,13 @@ def execute_model(
ctx = model_input.async_callback.keywords[ # type: ignore
"ctx"]
seq_group_metadata_list = ctx.seq_group_metadata_list
seq_group_metadata_list = copy.deepcopy(
seq_group_metadata_list)
# Cache the original output token ids
for i, seq_group_metadata in enumerate(
seq_group_metadata_list):
cache_orig_output_token_ids.append({})
for j, data in seq_group_metadata.seq_data.items():
cache_orig_output_token_ids[i][j] = \
copy.deepcopy(data.output_token_ids)
for seq_group_metadata in seq_group_metadata_list:
for data in seq_group_metadata.seq_data.values():
max_output_len = sampling_metadata.seq_groups[
Expand All @@ -2185,6 +2191,14 @@ def execute_model(
"attn_metadata":
self.trim_attn_metadata(result.attn_metadata)
})
else:
if len(cache_orig_output_token_ids) > 0:
# Reuse the original output token ids
for i, seq_group_metadata in enumerate(
seq_group_metadata_list):
for j, data in seq_group_metadata.seq_data.items():
data.output_token_ids = \
cache_orig_output_token_ids[i][j]

if self.is_driver_worker and self.profiler.enabled:
# Stop recording 'execute_model' event
Expand Down