DeepSeek/DeepEP
1
0
Fork 0
mirror of https://github.com/deepseek-ai/DeepEP synced 2025-06-26 18:28:11 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Support CUDA graph for intranode normal kernels (#203)

Browse Source
This commit is contained in:
Chenggang Zhao
2025-06-11 11:08:54 +08:00
committed by GitHub
parent 8da2d7b38d
commit a8299ca7c2
7 changed files with 86 additions and 38 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
csrc/deep_ep.hpp
View File

@@ -108,7 +108,8 @@ public:
const std::optional<torch::Tensor>& topk_idx, const std::optional<torch::Tensor>& topk_weights,
const std::optional<torch::Tensor>& num_tokens_per_rank, const torch::Tensor& is_token_in_rank, const std::optional<torch::Tensor>& num_tokens_per_expert,
int cached_num_recv_tokens, const std::optional<torch::Tensor>& cached_rank_prefix_matrix, const std::optional<torch::Tensor>& cached_channel_prefix_matrix,
int expert_alignment, const Config& config, std::optional<EventHandle>& previous_event, bool async, bool allocate_on_comm_stream);
int expert_alignment, int num_worst_tokens, const Config& config,
std::optional<EventHandle>& previous_event, bool async, bool allocate_on_comm_stream);
std::tuple<torch::Tensor, std::optional<torch::Tensor>, std::optional<EventHandle>>
intranode_combine(const torch::Tensor& x, const std::optional<torch::Tensor>& topk_weights,