Skip to content

Commit 1fb63c9

Browse files
NexesenexNexesenex
committed
Revert "iq1_kt: basics"
This reverts commit 3d635ed1820713f5f6a6a5cfb14993f73494a7c1.
1 parent 4b0fef9 commit 1fb63c9

17 files changed

Lines changed: 49 additions & 2978 deletions

File tree

‎ggml/include/ggml.h‎

Lines changed: 15 additions & 17 deletions
Original file line numberDiff line numberDiff line change
@@ -451,23 +451,21 @@ extern "C" {
451451
GGML_TYPE_IQ6_K = 141,
452452
// depricated: GGML_TYPE_IQ2_TN = 142,
453453
// depricated: GGML_TYPE_IQ1_TN = 143,
454-
455-
// IKL
456-
GGML_TYPE_IQ4_KS = 144,
457-
GGML_TYPE_IQ2_KS = 145,
458-
GGML_TYPE_IQ4_KSS = 146,
459-
GGML_TYPE_Q8_K16 = 147,
460-
GGML_TYPE_Q8_K32 = 148,
461-
GGML_TYPE_Q8_KR8 = 149,
462-
GGML_TYPE_Q8_K128 = 150,
463-
GGML_TYPE_Q8_KV = 151,
464-
GGML_TYPE_IQ5_KS = 152,
465-
GGML_TYPE_IQ2_KT = 153,
466-
GGML_TYPE_IQ3_KT = 154,
467-
GGML_TYPE_IQ4_KT = 155,
468-
GGML_TYPE_IQ3_KS = 156,
469-
GGML_TYPE_IQ2_KL = 157,
470-
GGML_TYPE_IQ1_KT = 158,
454+
// GGML_TYPE_IQ4_KS = 144,
455+
// GGML_TYPE_IQ2_KS = 145,
456+
// GGML_TYPE_IQ4_KSS = 146,
457+
// GGML_TYPE_Q8_K16 = 147,
458+
// GGML_TYPE_Q8_K32 = 148,
459+
// GGML_TYPE_Q8_KR8 = 149,
460+
// GGML_TYPE_Q8_K128 = 150,
461+
// GGML_TYPE_Q8_KV = 151,
462+
// GGML_TYPE_IQ5_KS = 152,
463+
// GGML_TYPE_IQ2_KT = 153,
464+
// GGML_TYPE_IQ3_KT = 154,
465+
// GGML_TYPE_IQ4_KT = 155,
466+
// GGML_TYPE_IQ3_KS = 156,
467+
// GGML_TYPE_IQ2_KL = 157,
468+
// GGML_TYPE_IQ1_KT = 158,
471469

472470
// GGML_TYPE_IQ3_KS_V1 = 196,
473471

‎ggml/src/ggml-cpu/ggml-cpu.c‎

Lines changed: 0 additions & 280 deletions
Original file line numberDiff line numberDiff line change
@@ -409,286 +409,6 @@ static const struct ggml_type_traits_cpu type_traits_cpu[GGML_TYPE_COUNT] = {
409409
.vec_dot_type = GGML_TYPE_BF16,
410410
.nrows = 1,
411411
},
412-
413-
// IKL
414-
[GGML_TYPE_BF16_R16] = {
415-
.from_float = (ggml_from_float_t) ggml_fp32_to_bf16_row,
416-
.vec_dot = (ggml_vec_dot_t) ggml_vec_dot_bf16,
417-
.vec_dot_type = GGML_TYPE_BF16,
418-
.nrows = 1,
419-
},
420-
/* [GGML_TYPE_Q4_0_4_4] = {
421-
.type_name = "q4_0_4x4",
422-
.blck_size = QK4_0,
423-
.blck_size_interleave = 4,
424-
.type_size = sizeof(block_q4_0),
425-
.is_quantized = true,
426-
.to_float = NULL,
427-
.from_float = NULL,
428-
.from_float_ref = NULL,
429-
.vec_dot = NULL,
430-
.vec_dot_type = GGML_TYPE_Q8_0,
431-
.nrows = 1,
432-
.ncols = 4,
433-
.gemv = ggml_gemv_q4_0_4x4_q8_0,
434-
.gemm = ggml_gemm_q4_0_4x4_q8_0,
435-
.row_meta_size = 0,
436-
},
437-
[GGML_TYPE_Q4_0_4_8] = {
438-
.type_name = "q4_0_4x8",
439-
.blck_size = QK4_0,
440-
.blck_size_interleave = 8,
441-
.type_size = sizeof(block_q4_0),
442-
.is_quantized = true,
443-
.to_float = NULL,
444-
.from_float = NULL,
445-
.from_float_ref = NULL,
446-
.vec_dot = NULL,
447-
.vec_dot_type = GGML_TYPE_Q8_0,
448-
.nrows = 1,
449-
.ncols = 4,
450-
.gemv = ggml_gemv_q4_0_4x8_q8_0,
451-
.gemm = ggml_gemm_q4_0_4x8_q8_0,
452-
.row_meta_size = 0,
453-
},
454-
[GGML_TYPE_Q4_0_8_8] = {
455-
.type_name = "q4_0_8x8",
456-
.blck_size = QK4_0,
457-
.blck_size_interleave = 8,
458-
.type_size = sizeof(block_q4_0),
459-
.is_quantized = true,
460-
.to_float = NULL,
461-
.from_float = NULL,
462-
.from_float_ref = NULL,
463-
.vec_dot = NULL,
464-
.vec_dot_type = GGML_TYPE_Q8_0,
465-
.nrows = 1,
466-
.ncols = 8,
467-
.gemv = ggml_gemv_q4_0_8x8_q8_0,
468-
.gemm = ggml_gemm_q4_0_8x8_q8_0,
469-
.row_meta_size = 0,
470-
}, */
471-
[GGML_TYPE_IQ2_K] = {
472-
.from_float = quantize_row_iq2_k,
473-
.vec_dot = vec_dot_iq2_k_q8_k,
474-
.vec_dot_type = GGML_TYPE_Q8_K,
475-
.nrows = 1,
476-
},
477-
[GGML_TYPE_IQ2_K_R4] = {
478-
.from_float = quantize_row_iq2_k_r4,
479-
.vec_dot = vec_dot_iq2_k_r4_q8_k,
480-
.vec_dot_type = GGML_TYPE_Q8_K,
481-
.nrows = 1,
482-
},
483-
[GGML_TYPE_IQ2_KS] = {
484-
.from_float = quantize_row_iq2_ks,
485-
.vec_dot = vec_dot_iq2_ks_q8_k,
486-
.vec_dot_type = GGML_TYPE_Q8_K,
487-
.nrows = 1,
488-
},
489-
[GGML_TYPE_IQ1_KT] = {
490-
.from_float = quantize_row_iq1_kt,
491-
.vec_dot = vec_dot_iq1_kt_q8_k,
492-
#if defined __AVX2__
493-
.vec_dot_type = GGML_TYPE_Q8_2_X4,
494-
#else
495-
.vec_dot_type = GGML_TYPE_Q8_0_X4,
496-
#endif
497-
.nrows = 1,
498-
},
499-
[GGML_TYPE_IQ2_KT] = {
500-
.from_float = quantize_row_iq2_kt,
501-
.vec_dot = vec_dot_iq2_kt_q8_k,
502-
#if defined __AVX2__
503-
.vec_dot_type = GGML_TYPE_Q8_2_X4,
504-
#else
505-
.vec_dot_type = GGML_TYPE_Q8_0_X4,
506-
#endif
507-
//#ifdef __ARM_NEON
508-
// .vec_dot_type = GGML_TYPE_F16,
509-
//#else
510-
// .vec_dot_type = GGML_TYPE_F32,
511-
//#endif
512-
.nrows = 1,
513-
},
514-
[GGML_TYPE_IQ3_KT] = {
515-
.from_float = quantize_row_iq3_kt,
516-
.vec_dot = vec_dot_iq3_kt_q8_k,
517-
#if defined __AVX2__
518-
.vec_dot_type = GGML_TYPE_Q8_2_X4,
519-
#else
520-
.vec_dot_type = GGML_TYPE_Q8_0_X4,
521-
#endif
522-
//#ifdef __ARM_NEON
523-
// .vec_dot_type = GGML_TYPE_F16,
524-
//#else
525-
// .vec_dot_type = GGML_TYPE_F32,
526-
//#endif
527-
.nrows = 1,
528-
},
529-
[GGML_TYPE_IQ4_KT] = {
530-
.from_float = quantize_row_iq4_kt,
531-
.vec_dot = vec_dot_iq4_kt_q8_k,
532-
#if defined __AVX2__
533-
.vec_dot_type = GGML_TYPE_Q8_2_X4,
534-
#else
535-
.vec_dot_type = GGML_TYPE_Q8_0_X4,
536-
#endif
537-
// #ifdef __ARM_NEON
538-
// .vec_dot_type = GGML_TYPE_F16,
539-
// #else
540-
// .vec_dot_type = GGML_TYPE_F32,
541-
// #endif
542-
.nrows = 1,
543-
},
544-
[GGML_TYPE_IQ3_K] = {
545-
.from_float = quantize_row_iq3_k,
546-
.vec_dot = vec_dot_iq3_k_q8_k,
547-
.vec_dot_type = GGML_TYPE_Q8_K,
548-
.nrows = 1,
549-
},
550-
[GGML_TYPE_IQ3_KS_V1] = {
551-
.from_float = quantize_row_iq3_ks_v1,
552-
.vec_dot = vec_dot_iq3_ks_v1_q8_k,
553-
.vec_dot_type = GGML_TYPE_Q8_K,
554-
.nrows = 1,
555-
},
556-
[GGML_TYPE_IQ3_KS] = {
557-
.from_float = quantize_row_iq3_ks,
558-
.vec_dot = vec_dot_iq3_ks_q8_k,
559-
.vec_dot_type = GGML_TYPE_Q8_K,
560-
.nrows = 1,
561-
},
562-
[GGML_TYPE_IQ2_KL] = {
563-
.from_float = quantize_row_iq2_kl,
564-
.vec_dot = vec_dot_iq2_kl_q8_k,
565-
.vec_dot_type = GGML_TYPE_Q8_K,
566-
.nrows = 1,
567-
},
568-
[GGML_TYPE_IQ4_K] = {
569-
.from_float = quantize_row_iq4_k,
570-
.vec_dot = vec_dot_iq4_k_q8_k,
571-
.vec_dot_type = GGML_TYPE_Q8_K,
572-
.nrows = 1,
573-
},
574-
[GGML_TYPE_IQ4_K_R4] = {
575-
.from_float = quantize_row_iq4_k_r4,
576-
.vec_dot = vec_dot_iq4_k_r4_q8_k,
577-
.vec_dot_type = GGML_TYPE_Q8_K,
578-
.nrows = 1,
579-
},
580-
[GGML_TYPE_IQ3_K_R4] = {
581-
.from_float = quantize_row_iq3_k_r4,
582-
.vec_dot = vec_dot_iq3_k_r4_q8_k,
583-
.vec_dot_type = GGML_TYPE_Q8_K,
584-
.nrows = 1,
585-
},
586-
[GGML_TYPE_IQ5_K] = {
587-
.from_float = quantize_row_iq5_k,
588-
.vec_dot = vec_dot_iq5_k_q8_k,
589-
// #ifdef __AVX2__
590-
// .vec_dot_type = GGML_TYPE_Q8_2_X4,
591-
// #else
592-
.vec_dot_type = GGML_TYPE_Q8_K,
593-
// #endif
594-
.nrows = 1,
595-
},
596-
[GGML_TYPE_IQ5_K_R4] = {
597-
.from_float = quantize_row_iq5_k_r4,
598-
.vec_dot = vec_dot_iq5_k_r4_q8_k,
599-
.vec_dot_type = GGML_TYPE_Q8_K,
600-
.nrows = 1,
601-
},
602-
[GGML_TYPE_IQ6_K] = {
603-
.from_float = quantize_row_iq6_k,
604-
.vec_dot = vec_dot_iq6_k_q8_k,
605-
.vec_dot_type = GGML_TYPE_Q8_K,
606-
.nrows = 1,
607-
},
608-
[GGML_TYPE_IQ4_NL_R4] = {
609-
.from_float = quantize_row_iq4_nl_r4,
610-
.vec_dot = vec_dot_iq4_nl_r4_q8_0,
611-
#ifdef GGML_USE_IQK_MULMAT
612-
#if defined __AVX2__
613-
.vec_dot_type = GGML_TYPE_Q8_2_X4,
614-
#else
615-
.vec_dot_type = GGML_TYPE_Q8_0_X4,
616-
#endif
617-
#else
618-
.vec_dot_type = GGML_TYPE_Q8_0,
619-
#endif
620-
.nrows = 1,
621-
},
622-
[GGML_TYPE_IQ4_XS_R8] = {
623-
.from_float = quantize_row_iq4_xs_r8,
624-
.vec_dot = vec_dot_iq4_xs_r8_q8_k,
625-
.vec_dot_type = GGML_TYPE_Q8_K32,
626-
.nrows = 1,
627-
},
628-
[GGML_TYPE_Q4_0_R8] = {
629-
.from_float = quantize_row_q4_0_r8,
630-
.vec_dot = vec_dot_q4_0_r8_q8_0,
631-
#ifdef GGML_USE_IQK_MULMAT
632-
#if defined __AVX2__
633-
.vec_dot_type = GGML_TYPE_Q8_2_X4,
634-
#else
635-
.vec_dot_type = GGML_TYPE_Q8_0_X4,
636-
#endif
637-
#else
638-
.vec_dot_type = GGML_TYPE_Q8_0,
639-
#endif
640-
.nrows = 1,
641-
},
642-
[GGML_TYPE_Q8_0_R8] = {
643-
.from_float = quantize_row_q8_0_r8,
644-
.vec_dot = vec_dot_q8_0_r8_q8_0,
645-
#ifdef GGML_USE_IQK_MULMAT
646-
#if defined __AVX2__
647-
.vec_dot_type = GGML_TYPE_Q8_2_X4,
648-
#else
649-
.vec_dot_type = GGML_TYPE_Q8_0_X4,
650-
#endif
651-
#else
652-
.vec_dot_type = GGML_TYPE_Q8_0,
653-
#endif
654-
.nrows = 1,
655-
},
656-
[GGML_TYPE_Q5_0_R4] = {
657-
.from_float = quantize_row_q5_0_r4,
658-
.vec_dot = vec_dot_q5_0_r4_q8_0,
659-
#ifdef GGML_USE_IQK_MULMAT
660-
#if defined __AVX2__
661-
.vec_dot_type = GGML_TYPE_Q8_2_X4,
662-
#else
663-
.vec_dot_type = GGML_TYPE_Q8_0_X4,
664-
#endif
665-
#else
666-
.vec_dot_type = GGML_TYPE_Q8_0,
667-
#endif
668-
.nrows = 1,
669-
},
670-
[GGML_TYPE_Q6_0_R4] = {
671-
.from_float = quantize_row_q6_0_r4,
672-
.vec_dot = vec_dot_q6_0_r4_q8_0,
673-
#ifdef GGML_USE_IQK_MULMAT
674-
#if defined __AVX2__
675-
.vec_dot_type = GGML_TYPE_Q8_2_X4,
676-
#else
677-
.vec_dot_type = GGML_TYPE_Q8_0_X4,
678-
#endif
679-
#else
680-
.vec_dot_type = GGML_TYPE_Q8_0,
681-
#endif
682-
.nrows = 1,
683-
},
684-
[GGML_TYPE_I2_S] = {
685-
.from_float = NULL,
686-
.vec_dot = NULL,
687-
.vec_dot_type = GGML_TYPE_Q8_0,
688-
.nrows = 1,
689-
},
690-
691-
// LCPP AGAIN
692412
[GGML_TYPE_TQ1_0] = {
693413
.from_float = quantize_row_tq1_0,
694414
.vec_dot = ggml_vec_dot_tq1_0_q8_K,

0 commit comments

Comments
 (0)