mirror of
https://github.com/PaddlePaddle/FastDeploy.git
synced 2025-12-24 13:28:13 +08:00
[XPU] Enable XPU V1 mode based on environment variable (#4213)
* Enable XPU V1 mode based on environment variable * add default param to xft_moe_fc_block_eb for latest xvllm compatibility; update run_ci_xpu to use latest xvllm
This commit is contained in:
@@ -72,6 +72,7 @@ void MoeExpertFFNImpl(xftblock::Tensor* ffn_in,
|
||||
is_padding_input ? token_num_info : nullptr,
|
||||
expert_num,
|
||||
1, // moe_topk
|
||||
0, // group_size
|
||||
ffn1_out_shape.size() == 2 ? xftblock::MoeFCInputMode::DENSE
|
||||
: xftblock::MoeFCInputMode::SPARSE);
|
||||
PD_CHECK(ret == 0);
|
||||
@@ -134,6 +135,7 @@ void MoeExpertFFNImpl(xftblock::Tensor* ffn_in,
|
||||
is_padding_input ? token_num_info : nullptr,
|
||||
expert_num,
|
||||
1, // moe_topk
|
||||
0, // group_size
|
||||
ffn1_out_shape.size() == 2
|
||||
? xftblock::MoeFCInputMode::DENSE
|
||||
: xftblock::MoeFCInputMode::SPARSE); // bias_mode
|
||||
|
||||
Reference in New Issue
Block a user