@@ -462,12 +462,9 @@ void quantize_row_q2_K(const float * restrict x, void * restrict vy, int k) {
462
462
}
463
463
464
464
size_t ggml_quantize_q2_K (const float * restrict src , void * restrict dst , int n , int k , int64_t * restrict hist ) {
465
- const int nb = k / QK_K ;
466
-
467
- // TODO - collect histograms - although, at a second thought, I don't really care about them
468
- (void )hist ;
465
+ (void )hist ; // TODO: collect histograms
469
466
470
- for (int j = 0 ; j < nb ; j += k ) {
467
+ for (int j = 0 ; j < n ; j += k ) {
471
468
block_q2_K * restrict y = (block_q2_K * )dst + j /QK_K ;
472
469
quantize_row_q2_K_reference (src + j , y , k );
473
470
}
@@ -678,12 +675,9 @@ void quantize_row_q3_K(const float * restrict x, void * restrict vy, int k) {
678
675
}
679
676
680
677
size_t ggml_quantize_q3_K (const float * restrict src , void * restrict dst , int n , int k , int64_t * restrict hist ) {
681
- const int nb = k / QK_K ;
682
-
683
- // TODO - collect histograms - although, at a second thought, I don't really care about them
684
- (void )hist ;
678
+ (void )hist ; // TODO: collect histograms
685
679
686
- for (int j = 0 ; j < nb ; j += k ) {
680
+ for (int j = 0 ; j < n ; j += k ) {
687
681
block_q3_K * restrict y = (block_q3_K * )dst + j /QK_K ;
688
682
quantize_row_q3_K_reference (src + j , y , k );
689
683
}
@@ -846,9 +840,9 @@ void quantize_row_q4_K(const float * restrict x, void * restrict vy, int k) {
846
840
847
841
size_t ggml_quantize_q4_K (const float * restrict src , void * restrict dst , int n , int k , int64_t * restrict hist ) {
848
842
assert (k % QK_K == 0 );
849
- const int nb = k / QK_K ;
850
843
(void )hist ; // TODO: collect histograms
851
- for (int j = 0 ; j < nb ; j += k ) {
844
+
845
+ for (int j = 0 ; j < n ; j += k ) {
852
846
block_q4_K * restrict y = (block_q4_K * )dst + j /QK_K ;
853
847
quantize_row_q4_K_reference (src + j , y , k );
854
848
}
@@ -1052,9 +1046,9 @@ void quantize_row_q5_K(const float * restrict x, void * restrict vy, int k) {
1052
1046
1053
1047
size_t ggml_quantize_q5_K (const float * restrict src , void * restrict dst , int n , int k , int64_t * restrict hist ) {
1054
1048
assert (k % QK_K == 0 );
1055
- const int nb = k / QK_K ;
1056
- ( void ) hist ;
1057
- for (int j = 0 ; j < nb ; j += k ) {
1049
+ ( void ) hist ; // TODO: collect histograms
1050
+
1051
+ for (int j = 0 ; j < n ; j += k ) {
1058
1052
block_q5_K * restrict y = (block_q5_K * )dst + j /QK_K ;
1059
1053
quantize_row_q5_K_reference (src + j , y , k );
1060
1054
}
@@ -1200,11 +1194,9 @@ void quantize_row_q6_K(const float * restrict x, void * restrict vy, int k) {
1200
1194
1201
1195
size_t ggml_quantize_q6_K (const float * src , void * dst , int n , int k , int64_t * hist ) {
1202
1196
assert (k % QK_K == 0 );
1203
- const int nb = k / QK_K ;
1204
-
1205
- (void )hist ; // TODO
1197
+ (void )hist ; // TODO: collect histograms
1206
1198
1207
- for (int j = 0 ; j < nb ; j += k ) {
1199
+ for (int j = 0 ; j < n ; j += k ) {
1208
1200
block_q6_K * restrict y = (block_q6_K * )dst + j /QK_K ;
1209
1201
quantize_row_q6_K_reference (src + j , y , k );
1210
1202
}
0 commit comments