v2: EmbeddingGemma multi-language support working (single request), MPS GPU fault known issue
This commit is contained in:
@@ -1637,10 +1637,11 @@ kernel void matmul_f32(
|
||||
uint id [[thread_position_in_grid]]
|
||||
) {
|
||||
// Each thread computes one output element
|
||||
uint row = 0; // For single token, M=1
|
||||
uint col = id;
|
||||
uint total = M * N;
|
||||
if (id >= total) return;
|
||||
|
||||
if (col >= N) return;
|
||||
uint row = id / N;
|
||||
uint col = id % N;
|
||||
|
||||
float sum = 0.0;
|
||||
for (uint k = 0; k < K; k++) {
|
||||
|
||||
Reference in New Issue
Block a user