apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-12-24 13:28:13 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[UNITEST] add test (#5305)

Browse Source
This commit is contained in:
周周周
2025-12-02 17:59:01 +08:00
committed by GitHub
parent 8e0f4dfd0c
commit fb7f951612
3 changed files with 114 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
custom_ops/gpu_ops/moe/ep_moe_expert_dispatch.cu
View File

@@ -847,8 +847,11 @@ __global__ void permute_x_fp8_kernel(
const int start_idx = i == 0 ? 0 : token_nums_per_expert_cum[i - 1];
const int end_idx = token_nums_per_expert_cum[i];
if (s_token_idx >= start_idx && s_token_idx < end_idx) {
if ((s_token_idx - start_idx) < token_nums_per_expert[i])
if ((s_token_idx - start_idx) < token_nums_per_expert[i]) {
m_indices[s_token_idx] = i;
} else {
m_indices[s_token_idx] = -1;
}
break;
}
}
@@ -984,8 +987,8 @@ std::vector<paddle::Tensor> EPMoeExpertDispatchFP8(
paddle::DataType::FLOAT32,
place);
auto m_indices = paddle::full(
{token_nums_feed_to_ffn}, -1, paddle::DataType::INT32, place);
auto m_indices =
GetEmptyTensor({token_nums_feed_to_ffn}, paddle::DataType::INT32, place);
auto token_nums_per_expert_cumsum =
GetEmptyTensor({num_experts_per_rank}, paddle::DataType::INT64, place);
auto token_nums_per_expert_padded_cumsum =