fix(build): correct preprocessor guard for GDN decode to fix Ampere c… (#52743)
Signed-off-by: prakharP <prakharpant288@gmail.com>
This commit is contained in:
@@ -393,6 +393,9 @@ void fused_kda_decode(
|
|||||||
std::optional<torch::stable::Tensor> output_gate,
|
std::optional<torch::stable::Tensor> output_gate,
|
||||||
std::optional<torch::stable::Tensor> norm_weight, double norm_eps);
|
std::optional<torch::stable::Tensor> norm_weight, double norm_eps);
|
||||||
|
|
||||||
|
#endif
|
||||||
|
|
||||||
|
#ifdef VLLM_ENABLE_FUSED_GDN_DECODE
|
||||||
void fused_gdn_decode_post_conv_mtp(
|
void fused_gdn_decode_post_conv_mtp(
|
||||||
torch::stable::Tensor const& mixed_qkv, torch::stable::Tensor const& a,
|
torch::stable::Tensor const& mixed_qkv, torch::stable::Tensor const& a,
|
||||||
torch::stable::Tensor const& b, torch::stable::Tensor const& a_log,
|
torch::stable::Tensor const& b, torch::stable::Tensor const& a_log,
|
||||||
|
|||||||
Reference in New Issue
Block a user