k-quants : support for super-block size of 64 (#2001)

* k_quants: WIP super-blocks with 64 weights

* k_quants: WIP super-blocks with 64 weights

Q6_K scalar and AVX2 works

* k_quants: WIP super-blocks with 64 weights

Q4_K scalar and AVX2 works

* k_quants: WIP super-blocks with 64 weights

Q2_K scalar and AVX2 works. Q2_K is way too slow (it is actually slower
than the scalar implementation)

* k_quants: WIP super-blocks with 64 weights

Q3_K scalar and AVX2 works.

* k_quants: WIP super-blocks with 64 weights

Q5_K scalar and AVX2 works, and with that all
k_quants are done on AVX2 and scalar

* k_quants: WIP super-blocks with 64 weights

Q6_K working on CUDA. Cannot make it run quite as gast as
with super-blocks with 256 weigths: 8% slower on 4080,
20% slower on the 1660 (but there we fit 1 less layer on the
GPU because pf the larger model size), so some fraction of
these 20% is due to that,

* k_quants: WIP super-blocks with 64 weights

Q4_K working on CUDA. ~10% slower on GTX-1660,
16% slower on 4080.

* k_quants: WIP super-blocks with 64 weights

Q2_K working on CUDA. ~3% slower on GTX-1660,
10% slower on 4080.

* k_quants: WIP super-blocks with 64 weights

Q3_K working on CUDA.

* k_quants: WIP super-blocks with 64 weights

Q5_K working on CUDA, and with this CUDA is done.

* k_quants: WIP super-blocks with 64 weights

Q6_K working on ARM_NEON

* k_quants: WIP super-blocks with 64 weights

Q4_K working on ARM_NEON, but quite a bit slower than 256 weights

* k_quants: WIP super-blocks with 64 weights

Q2_K working on ARM_NEON, but quite a bit slower than 256 weights

* k_quants: WIP super-blocks with 64 weights

Q3_K working on ARM_NEON, but quite a bit slower than 256 weights.

* k_quants: WIP super-blocks with 64 weights

Q5_K working on ARM_NEON, but quite a bit slower than 256 weights.

With that, we have full support for ARM_NEON, although
performance is not quite there.

* k_quants: WIP super-blocks with 64 weights

Slightly more efficient Q3_K and Q5_K

* k_quants: WIP super-blocks with 64 weights

Another small improvement for Q3_K and Q5_K on ARM_NEON

* k_quants: WIP super-blocks with 64 weights

Yet another speedup for Q5_K on ARM_NEON.
We are now within 10% of the QK_K = 256 version.

* k_quants: WIP super-blocks with 64 weights

* We are able to pass preprocessor macros to the Metal
  compiler
* Q6_K works and is actually slightly more efficient than
  the QK_K = 256 version (25.2 ms vs 25.8 ms)

* k_quants: WIP super-blocks with 64 weights

Q4_K works on Metal and is actually slightly faster
than QK_K = 256 (21.95 ms vs 24.0 ms).

* k_quants: WIP super-blocks with 64 weights

Q2_K works on Metal and is very slightly faster
than QK_K = 256 (23.8 ms vs 24.2 ms).

* k_quants: WIP super-blocks with 64 weights

Q3_K works on Metal and is slightly faster
than QK_K = 256 (26.6 ms vs 28.3 ms).

* k_quants: WIP super-blocks with 64 weights

Q5_K works on Metal and is slightly faster
than QK_K = 256 (23.7 ms vs 26.3 ms).

* k_quants: call them _K, not _k, also on Metal

* k_quants: correctly define QK_K in llama.cpp

* Fixed bug in q4_K quantization added with the 64-block addition

* Simplify via lambda

* k_quants: swicth Q3_K to 4-bit scales when QK_K = 64

Otherwise there isn't much benefit from this
quantization type. There is some very slight loss
in accuracy, but we reduce size by ~7%.
E.g., for OpenLLaMA-3B, Q3_K_S perplexity is
8.6131 with 8-bit scales and 8.6352 with 4-bit,
while file size decreases from 1.53G to 1.44G.

* k_quants: switch Q4_K to 4-bit scales when QK_K = 64

 Here the loss in accuracy is greater than for Q3_K,
 but the Q4_K points still move further to the left on
 the perplexity vs size curve.

* k_quants: forgot to add the Metal changes in last commit

* k_quants: change Q5_K to be type 0 when QK_K = 64

Still needs AVX2 implementation

* k_quants: AVX2 implementation for new 64-weight Q5_K

* k_quants: 10% faster ARM_NEON Q5_K dot product

* k_quants: fixed issue caused by merging with master

---------

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>

1 files changed, 41 insertions, 6 deletions

diff --git a/k_quants.h b/k_quants.h
index 10a0baa..6abe3d7 100644
--- a/k_quants.h
+++ b/k_quants.h
@@ -7,7 +7,13 @@
 #include <stddef.h>
 
 // Super-block size
+#ifdef GGML_QKK_64
+#define QK_K 64
+#define K_SCALE_SIZE 4
+#else
 #define QK_K 256
+#define K_SCALE_SIZE 12
+#endif
 
 //
 // Super-block quantization structures
@@ -29,38 +35,67 @@ static_assert(sizeof(block_q2_K) == 2*sizeof(ggml_fp16_t) + QK_K/16 + QK_K/4, "w
 // weight is represented as x = a * q
 // 16 blocks of 16 elemenets each
 // Effectively 3.4375 bits per weight
+#ifdef GGML_QKK_64
 typedef struct {
     uint8_t hmask[QK_K/8];     // quants - high bit
     uint8_t qs[QK_K/4];        // quants - low 2 bits
-    uint8_t scales[3*QK_K/64]; // scales, quantized with 6 bits
+    uint8_t scales[2];
     ggml_fp16_t d;             // super-block scale
 } block_q3_K;
-static_assert(sizeof(block_q3_K) == sizeof(ggml_fp16_t) + QK_K / 4 + 11 * QK_K / 64, "wrong q3_K block size/padding");
+static_assert(sizeof(block_q3_K) == sizeof(ggml_fp16_t) + QK_K / 4 + QK_K / 8 + 2, "wrong q3_K block size/padding");
+#else
+typedef struct {
+    uint8_t hmask[QK_K/8];     // quants - high bit
+    uint8_t qs[QK_K/4];        // quants - low 2 bits
+    uint8_t scales[12];        // scales, quantized with 6 bits
+    ggml_fp16_t d;             // super-block scale
+} block_q3_K;
+static_assert(sizeof(block_q3_K) == sizeof(ggml_fp16_t) + QK_K / 4 + QK_K / 8 + 12, "wrong q3_K block size/padding");
+#endif
 
 // 4-bit quantization
 // 16 blocks of 32 elements each
 // weight is represented as x = a * q + b
 // Effectively 4.5 bits per weight
+#ifdef GGML_QKK_64
+typedef struct {
+    ggml_fp16_t d[2];          // super-block scales/mins
+    uint8_t scales[2];         // 4-bit block scales/mins
+    uint8_t qs[QK_K/2];        // 4--bit quants
+} block_q4_K;
+static_assert(sizeof(block_q4_K) == 2*sizeof(ggml_fp16_t) + QK_K/2 + 2, "wrong q4_K block size/padding");
+#else
 typedef struct {
     ggml_fp16_t d;             // super-block scale for quantized scales
     ggml_fp16_t dmin;          // super-block scale for quantized mins
-    uint8_t scales[3*QK_K/64]; // scales and mins, quantized with 6 bits
+    uint8_t scales[K_SCALE_SIZE]; // scales and mins, quantized with 6 bits
     uint8_t qs[QK_K/2];        // 4--bit quants
 } block_q4_K;
-static_assert(sizeof(block_q4_K) == 2*sizeof(ggml_fp16_t) + 3*QK_K/64 + QK_K/2, "wrong q4_K block size/padding");
+static_assert(sizeof(block_q4_K) == 2*sizeof(ggml_fp16_t) + K_SCALE_SIZE + QK_K/2, "wrong q4_K block size/padding");
+#endif
 
 // 5-bit quantization
 // 16 blocks of 32 elements each
 // weight is represented as x = a * q + b
 // Effectively 5.5 bits per weight
+#ifdef GGML_QKK_64
+typedef struct {
+    ggml_fp16_t d;               // super-block scale
+    int8_t  scales[QK_K/16];     // 8-bit block scales
+    uint8_t qh[QK_K/8];          // quants, high bit
+    uint8_t qs[QK_K/2];          // quants, low 4 bits
+} block_q5_K;
+static_assert(sizeof(block_q5_K) == sizeof(ggml_fp16_t) + QK_K/2 + QK_K/8 + QK_K/16, "wrong q5_K block size/padding");
+#else
 typedef struct {
     ggml_fp16_t d;               // super-block scale for quantized scales
     ggml_fp16_t dmin;            // super-block scale for quantized mins
-    uint8_t scales[3*QK_K/64];   // scales and mins, quantized with 6 bits
+    uint8_t scales[K_SCALE_SIZE];   // scales and mins, quantized with 6 bits
     uint8_t qh[QK_K/8];          // quants, high bit
     uint8_t qs[QK_K/2];          // quants, low 4 bits
 } block_q5_K;
-static_assert(sizeof(block_q5_K) == 2*sizeof(ggml_fp16_t) + 3*QK_K/64 + QK_K/2 + QK_K/8, "wrong q5_K block size/padding");
+static_assert(sizeof(block_q5_K) == 2*sizeof(ggml_fp16_t) + K_SCALE_SIZE + QK_K/2 + QK_K/8, "wrong q5_K block size/padding");
+#endif
 
 // 6-bit quantization
 // weight is represented as x = a * q