Skip to content

fix: preserve multi_modal_data in generate_opt_level=0 path #446

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

Open
sanmuf wants to merge 1 commit into
base: main
Choose a base branch
from
Open

fix: preserve multi_modal_data in generate_opt_level=0 path #446

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
11 changes: 10 additions & 1 deletion roll/distributed/scheduler/generate_scheduler.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -567,7 +567,16 @@ async def get_batch_opt_level_0(self, data: DataProto, batch_size: int) -> DataP
request_data: DataProto = DataProto.from_single_dict(collect_data, meta_info=data.meta_info)
request_data.batch["prompt_id"] = torch.arange(request_data.batch.batch_size[0], device=request_data.batch.device)

gen_batch = request_data.pop(batch_keys=["input_ids", "attention_mask", "position_ids"])
generate_non_tensor_batch_keys = []
if "multi_modal_data" in request_data.non_tensor_batch:
generate_non_tensor_batch_keys.append("multi_modal_data")

gen_batch = request_data.pop(
batch_keys=["input_ids", "attention_mask", "position_ids"],
non_tensor_batch_keys=generate_non_tensor_batch_keys,
)
for key in generate_non_tensor_batch_keys:
request_data.non_tensor_batch[key] = gen_batch.non_tensor_batch[key]
gen_batch.meta_info = request_data.meta_info
num_return_sequences = generation_config["num_return_sequences"]
request_data = request_data.repeat(repeat_times=num_return_sequences)
Expand Down