vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

remove block scale

Browse source
This commit is contained in:
Eddie-Wang1120 2024-06-18 23:33:58 +08:00
parent 4edc958fec
commit 89c7e4c1dd
3 changed files with 3 additions and 11 deletions
Download patch file Download diff file Expand all files Collapse all files

2
examples/quantize/quantize.cpp
View file

@ -26,7 +26,7 @@ static const std::vector<struct quant_option> QUANT_OPTIONS = {
{ "IQ2_M", LLAMA_FTYPE_MOSTLY_IQ2_M, " 2.7 bpw quantization", },
{ "IQ1_S", LLAMA_FTYPE_MOSTLY_IQ1_S, " 1.56 bpw quantization", },
{ "IQ1_M", LLAMA_FTYPE_MOSTLY_IQ1_M, " 1.75 bpw quantization", },
{ "Q2_2", LLAMA_FTYPE_MOSTLY_Q2_2, " 2.5 bpw quantization", },
{ "Q2_2", LLAMA_FTYPE_MOSTLY_Q2_2, " 2 bpw quantization", },
{ "Q2_K", LLAMA_FTYPE_MOSTLY_Q2_K, " 2.63G, +0.6717 ppl @ LLaMA-v1-7B", },
{ "Q2_K_S", LLAMA_FTYPE_MOSTLY_Q2_K_S, " 2.16G, +9.0634 ppl @ LLaMA-v1-7B", },
{ "IQ3_XXS",LLAMA_FTYPE_MOSTLY_IQ3_XXS," 3.06 bpw quantization", },

3
ggml-common.h
View file

@ -139,10 +139,9 @@ typedef sycl::half2 ggml_half2;
#define QK2_2 32
typedef struct {
ggml_half d; // delta
uint8_t qs[QK2_2 / 4]; // nibbles / quants
} block_q2_2;
static_assert(sizeof(block_q2_2) == sizeof(ggml_half) + QK2_2 / 4, "wrong q2_2 block size/padding");
static_assert(sizeof(block_q2_2) == QK2_2 / 4, "wrong q2_2 block size/padding");
#define QK4_0 32
typedef struct {

9
ggml-quants.c
View file

@ -668,10 +668,6 @@ void quantize_row_q2_2_reference(const float * restrict x, block_q2_2 * restrict
for (int i = 0; i < nb; i++) {
const float d = 1.0f;
y[i].d = GGML_FP32_TO_FP16(d);
for (int j = 0; j < qk/4; ++j) {
int8_t x0 = (int8_t)x[i*qk + j*4 + 0];
int8_t x1 = (int8_t)x[i*qk + j*4 + 1];
@ -14369,10 +14365,6 @@ bool ggml_validate_row_data(enum ggml_type type, const void * data, size_t nbyte
}
}
} break;
case GGML_TYPE_Q2_2:
{
VALIDATE_ROW_DATA_D_F16_IMPL(block_q2_2, data, nb);
} break;
case GGML_TYPE_Q4_0:
{
VALIDATE_ROW_DATA_D_F16_IMPL(block_q4_0, data, nb);
@ -14467,6 +14459,7 @@ bool ggml_validate_row_data(enum ggml_type type, const void * data, size_t nbyte
{
VALIDATE_ROW_DATA_D_F16_IMPL(block_iq4_nl, data, nb);
} break;
case GGML_TYPE_Q2_2:
case GGML_TYPE_I8:
case GGML_TYPE_I16:
case GGML_TYPE_I32: