Skip to content

[V1][Minor] Optimize token_ids_cpu copy #11692

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 1 commit into from
Jan 2, 2025
Merged

[V1][Minor] Optimize token_ids_cpu copy #11692

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
13 changes: 8 additions & 5 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 @@ -66,8 +66,9 @@ def __init__(
pin_memory=False,
)
self.token_ids_cpu = self.token_ids_cpu_tensor.numpy()
self.num_computed_tokens_cpu = np.empty(max_num_reqs, dtype=np.int32)
self.num_tokens = np.zeros(max_num_reqs, dtype=np.int32)
self.num_prompt_tokens = np.zeros(max_num_reqs, dtype=np.int32)
self.num_computed_tokens_cpu = np.empty(max_num_reqs, dtype=np.int32)

# Attention-related.
self.block_table = torch.zeros(
Expand Down Expand Up @@ -189,6 +190,7 @@ def add_request(
end_idx = start_idx + len(request.output_token_ids)
self.token_ids_cpu[req_index,
start_idx:end_idx] = request.output_token_ids
self.num_tokens[req_index] = request.num_tokens

self.num_computed_tokens_cpu[req_index] = request.num_computed_tokens
num_blocks = len(request.block_ids)
Expand Down Expand Up @@ -290,14 +292,15 @@ def condense(self, empty_req_indices: List[int]) -> None:
self.req_ids[last_req_index] = None
self.req_id_to_index[req_id] = empty_index

# TODO(woosuk): Optimize the copy of token_ids_cpu and
# block_table_cpu.
self.token_ids_cpu[empty_index] = self.token_ids_cpu[
last_req_index]
num_tokens = self.num_tokens[last_req_index]
self.token_ids_cpu[empty_index, :num_tokens] = self.token_ids_cpu[
last_req_index, :num_tokens]
self.num_tokens[empty_index] = num_tokens
self.num_prompt_tokens[empty_index] = \
self.num_prompt_tokens[last_req_index]
self.num_computed_tokens_cpu[
empty_index] = self.num_computed_tokens_cpu[last_req_index]
# TODO(woosuk): Optimize the copy of block_table_cpu.
self.block_table_cpu[empty_index] = self.block_table_cpu[
last_req_index]
self.temperature_cpu[empty_index] = self.temperature_cpu[
Expand Down
1 change: 1 addition & 0 deletions vllm/v1/worker/gpu_model_runner.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -644,6 +644,7 @@ def execute_model(
# Append the sampled token to the output token ids.
token_id = sampled_token_ids[i]
self.input_batch.token_ids_cpu[i, seq_len] = token_id
self.input_batch.num_tokens[i] += 1
req_state.output_token_ids.append(token_id)
else:
# Ignore the sampled token from the partial request.
Expand Down
Loading