mirror of
https://github.com/ggerganov/llama.cpp.git
synced 2024-12-24 13:28:50 +01:00
CUDA: fix matrix multiplication logic for tests (#6667)
This commit is contained in:
parent
4bd0f93e4a
commit
b5e7285baf
@ -1946,7 +1946,7 @@ static void ggml_cuda_mul_mat(ggml_backend_cuda_context & ctx, const ggml_tensor
|
||||
} else if (!split && !fp16_performance_good && src0->type == GGML_TYPE_F16 && !ggml_is_contiguous(src0) && !ggml_is_transposed(src1) && src1->ne[1] == 1) {
|
||||
// KQV single-batch
|
||||
ggml_cuda_mul_mat_vec_nc(ctx, src0, src1, dst);
|
||||
} else if (!split && fp16_performance_good && src0->type == GGML_TYPE_F16 && !ggml_is_transposed(src0) && !ggml_is_transposed(src1) && src1->ne[2]*src1->ne[3] > 1) {
|
||||
} else if (!split && src0->type == GGML_TYPE_F16 && (src1->type == GGML_TYPE_F16 || fp16_performance_good) && !ggml_is_transposed(src0) && !ggml_is_transposed(src1) && src1->ne[2]*src1->ne[3] > 1) {
|
||||
// KQ + KQV multi-batch
|
||||
ggml_cuda_mul_mat_batched_cublas(ctx, src0, src1, dst);
|
||||
} else if (use_dequantize_mul_mat_vec) {
|
||||
|
Loading…
Reference in New Issue
Block a user