[Quant][Kernel] Remove GPTQ Group/Dynamic Activation Ordering (#54809)
Signed-off-by: Roderick-Wu <roderickwu2003@gmail.com> Signed-off-by: Roderick Wu <roderickwu2003@gmail.com> Signed-off-by: mgoin <mgoin64@gmail.com> Co-authored-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
+2
-3
@@ -26,12 +26,11 @@ void wvSplitKQ(const at::Tensor& in_a, const at::Tensor& in_b,
|
||||
|
||||
torch::Tensor gptq_gemm_rdna3(torch::Tensor a, torch::Tensor b_q_weight,
|
||||
torch::Tensor b_qzeros, torch::Tensor b_scales,
|
||||
torch::Tensor b_g_idx, bool use_v2_format);
|
||||
bool use_v2_format);
|
||||
|
||||
torch::Tensor gptq_gemm_rdna3_wmma(torch::Tensor a, torch::Tensor b_q_weight,
|
||||
torch::Tensor b_qzeros,
|
||||
torch::Tensor b_scales,
|
||||
torch::Tensor b_g_idx, bool use_v2_format);
|
||||
torch::Tensor b_scales, bool use_v2_format);
|
||||
|
||||
void moe_gptq_gemm_rdna3(torch::Tensor a, torch::Tensor c,
|
||||
torch::Tensor b_q_weight, torch::Tensor b_scales,
|
||||
|
||||
Reference in New Issue
Block a user