Skip to content

Fix DeepEP cannot be used together with code that needs GIL such as Mooncake transfer engine #142

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
LyricZhao merged 2 commits into from
May 8, 2025
Merged

Fix DeepEP cannot be used together with code that needs GIL such as Mooncake transfer engine #142

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
23ded3b
Update deep_ep.cpp
fzyzcjy Apr 29, 2025
adc6e24
Update deep_ep.cpp
fzyzcjy May 8, 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
5 changes: 5 additions & 0 deletions csrc/deep_ep.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -614,6 +614,11 @@ Buffer::internode_dispatch(const torch::Tensor& x, const std::optional<torch::Te
const std::optional<torch::Tensor>& cached_rdma_channel_prefix_matrix, const std::optional<torch::Tensor>& cached_recv_rdma_rank_prefix_sum,
const std::optional<torch::Tensor>& cached_gbl_channel_prefix_matrix, const std::optional<torch::Tensor>& cached_recv_gbl_rank_prefix_sum,
int expert_alignment, const Config& config, std::optional<EventHandle>& previous_event, bool async, bool allocate_on_comm_stream) {
// In dispatch, CPU will busy-wait until GPU receive tensor size metadata from other ranks, which can be quite long.
// If users of DeepEP need to execute other Python code on other threads, such as KV transfer, their code will get stuck due to GIL
// unless we release GIL here.
pybind11::gil_scoped_release release;

const int num_channels = config.num_sms / 2;
EP_HOST_ASSERT(config.num_sms % 2 == 0);
EP_HOST_ASSERT(0 < get_num_rdma_ranks() and get_num_rdma_ranks() <= NUM_MAX_RDMA_PEERS);
Expand Down