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 bias. (#257)

Browse Source 
* Support bias.

* Fix.

* Fix style.
This commit is contained in:
Shangyan Zhou
2025-06-25 13:04:20 +08:00
parent 85adc566e2
commit 4931324861
7 changed files with 101 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
csrc/kernels/api.cuh
View File

@@ -68,6 +68,7 @@ void cached_notify_combine(void** buffer_ptrs, int* send_head, int num_channels,
void combine(cudaDataType_t type,
void* recv_x, float* recv_topk_weights,
const void* x, const float* topk_weights,
const void* bias_0, const void* bias_1,
const int* src_idx, const int* rank_prefix_matrix, const int* channel_prefix_matrix,
int* send_head, int num_tokens, int num_recv_tokens, int hidden, int num_topk,
void** buffer_ptrs, int rank, int num_ranks,
@@ -121,6 +122,7 @@ void combine(cudaDataType_t type,
void* combined_x, float* combined_topk_weights,
const bool* is_combined_token_in_rank,
const void* x, const float* topk_weights,
const void* bias_0, const void* bias_1,
const int* combined_rdma_head, const int* combined_nvl_head,
const void* src_meta, const int* rdma_channel_prefix_matrix, const int* rdma_rank_prefix_sum, const int* gbl_channel_prefix_matrix,
int num_tokens, int num_combined_tokens, int hidden, int num_topk,