Merge branch 'master' into fix-refact

This commit is contained in:
Georgi Gerganov 2023-10-08 11:25:16 +03:00
commit acead654d2
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
12 changed files with 223 additions and 170 deletions

View file

@ -81,18 +81,18 @@ struct ggml_metal_context {
GGML_METAL_DECL_KERNEL(get_rows_q6_K);
GGML_METAL_DECL_KERNEL(rms_norm);
GGML_METAL_DECL_KERNEL(norm);
GGML_METAL_DECL_KERNEL(mul_mat_f32_f32);
GGML_METAL_DECL_KERNEL(mul_mat_f16_f32);
GGML_METAL_DECL_KERNEL(mul_mat_f16_f32_1row);
GGML_METAL_DECL_KERNEL(mul_mat_f16_f32_l4);
GGML_METAL_DECL_KERNEL(mul_mat_q4_0_f32);
GGML_METAL_DECL_KERNEL(mul_mat_q4_1_f32);
GGML_METAL_DECL_KERNEL(mul_mat_q8_0_f32);
GGML_METAL_DECL_KERNEL(mul_mat_q2_K_f32);
GGML_METAL_DECL_KERNEL(mul_mat_q3_K_f32);
GGML_METAL_DECL_KERNEL(mul_mat_q4_K_f32);
GGML_METAL_DECL_KERNEL(mul_mat_q5_K_f32);
GGML_METAL_DECL_KERNEL(mul_mat_q6_K_f32);
GGML_METAL_DECL_KERNEL(mul_mv_f32_f32);
GGML_METAL_DECL_KERNEL(mul_mv_f16_f32);
GGML_METAL_DECL_KERNEL(mul_mv_f16_f32_1row);
GGML_METAL_DECL_KERNEL(mul_mv_f16_f32_l4);
GGML_METAL_DECL_KERNEL(mul_mv_q4_0_f32);
GGML_METAL_DECL_KERNEL(mul_mv_q4_1_f32);
GGML_METAL_DECL_KERNEL(mul_mv_q8_0_f32);
GGML_METAL_DECL_KERNEL(mul_mv_q2_K_f32);
GGML_METAL_DECL_KERNEL(mul_mv_q3_K_f32);
GGML_METAL_DECL_KERNEL(mul_mv_q4_K_f32);
GGML_METAL_DECL_KERNEL(mul_mv_q5_K_f32);
GGML_METAL_DECL_KERNEL(mul_mv_q6_K_f32);
GGML_METAL_DECL_KERNEL(mul_mm_f32_f32);
GGML_METAL_DECL_KERNEL(mul_mm_f16_f32);
GGML_METAL_DECL_KERNEL(mul_mm_q4_0_f32);
@ -185,56 +185,44 @@ struct ggml_metal_context * ggml_metal_init(int n_cb) {
ctx->d_queue = dispatch_queue_create("ggml-metal", DISPATCH_QUEUE_CONCURRENT);
#ifdef GGML_SWIFT
// load the default.metallib file
// load library
{
NSError * error = nil;
NSBundle * bundle = [NSBundle bundleForClass:[GGMLMetalClass class]];
NSString * llamaBundlePath = [bundle pathForResource:@"llama_llama" ofType:@"bundle"];
NSBundle * llamaBundle = [NSBundle bundleWithPath:llamaBundlePath];
NSString * libPath = [llamaBundle pathForResource:@"default" ofType:@"metallib"];
NSURL * libURL = [NSURL fileURLWithPath:libPath];
// Load the metallib file into a Metal library
ctx->library = [ctx->device newLibraryWithURL:libURL error:&error];
if (error) {
GGML_METAL_LOG_ERROR("%s: error: %s\n", __func__, [[error description] UTF8String]);
return NULL;
}
}
NSBundle * bundle = nil;
#ifdef SWIFT_PACKAGE
bundle = SWIFTPM_MODULE_BUNDLE;
#else
UNUSED(msl_library_source);
// read the source from "ggml-metal.metal" into a string and use newLibraryWithSource
{
bundle = [NSBundle bundleForClass:[GGMLMetalClass class]];
#endif
NSError * error = nil;
NSString * libPath = [bundle pathForResource:@"default" ofType:@"metallib"];
if (libPath != nil) {
NSURL * libURL = [NSURL fileURLWithPath:libPath];
GGML_METAL_LOG_INFO("%s: loading '%s'\n", __func__, [libPath UTF8String]);
ctx->library = [ctx->device newLibraryWithURL:libURL error:&error];
} else {
GGML_METAL_LOG_INFO("%s: default.metallib not found, loading from source\n", __func__);
//NSString * path = [[NSBundle mainBundle] pathForResource:@"../../examples/metal/metal" ofType:@"metal"];
NSBundle * bundle = [NSBundle bundleForClass:[GGMLMetalClass class]];
NSString * path = [bundle pathForResource:@"ggml-metal" ofType:@"metal"];
GGML_METAL_LOG_INFO("%s: loading '%s'\n", __func__, [path UTF8String]);
NSString * src = [NSString stringWithContentsOfFile:path encoding:NSUTF8StringEncoding error:&error];
if (error) {
GGML_METAL_LOG_ERROR("%s: error: %s\n", __func__, [[error description] UTF8String]);
return NULL;
}
NSString * sourcePath = [bundle pathForResource:@"ggml-metal" ofType:@"metal"];
GGML_METAL_LOG_INFO("%s: loading '%s'\n", __func__, [sourcePath UTF8String]);
NSString * src = [NSString stringWithContentsOfFile:sourcePath encoding:NSUTF8StringEncoding error:&error];
if (error) {
GGML_METAL_LOG_ERROR("%s: error: %s\n", __func__, [[error description] UTF8String]);
return NULL;
}
MTLCompileOptions* options = nil;
#ifdef GGML_QKK_64
MTLCompileOptions* options = [MTLCompileOptions new];
options.preprocessorMacros = @{ @"QK_K" : @(64) };
ctx->library = [ctx->device newLibraryWithSource:src options:options error:&error];
#else
ctx->library = [ctx->device newLibraryWithSource:src options:nil error:&error];
options = [MTLCompileOptions new];
options.preprocessorMacros = @{ @"QK_K" : @(64) };
#endif
ctx->library = [ctx->device newLibraryWithSource:src options:options error:&error];
}
if (error) {
GGML_METAL_LOG_ERROR("%s: error: %s\n", __func__, [[error description] UTF8String]);
return NULL;
}
}
#endif
// load kernels
{
@ -274,28 +262,30 @@ struct ggml_metal_context * ggml_metal_init(int n_cb) {
GGML_METAL_ADD_KERNEL(get_rows_q6_K);
GGML_METAL_ADD_KERNEL(rms_norm);
GGML_METAL_ADD_KERNEL(norm);
GGML_METAL_ADD_KERNEL(mul_mat_f32_f32);
GGML_METAL_ADD_KERNEL(mul_mat_f16_f32);
GGML_METAL_ADD_KERNEL(mul_mat_f16_f32_1row);
GGML_METAL_ADD_KERNEL(mul_mat_f16_f32_l4);
GGML_METAL_ADD_KERNEL(mul_mat_q4_0_f32);
GGML_METAL_ADD_KERNEL(mul_mat_q4_1_f32);
GGML_METAL_ADD_KERNEL(mul_mat_q8_0_f32);
GGML_METAL_ADD_KERNEL(mul_mat_q2_K_f32);
GGML_METAL_ADD_KERNEL(mul_mat_q3_K_f32);
GGML_METAL_ADD_KERNEL(mul_mat_q4_K_f32);
GGML_METAL_ADD_KERNEL(mul_mat_q5_K_f32);
GGML_METAL_ADD_KERNEL(mul_mat_q6_K_f32);
GGML_METAL_ADD_KERNEL(mul_mm_f32_f32);
GGML_METAL_ADD_KERNEL(mul_mm_f16_f32);
GGML_METAL_ADD_KERNEL(mul_mm_q4_0_f32);
GGML_METAL_ADD_KERNEL(mul_mm_q8_0_f32);
GGML_METAL_ADD_KERNEL(mul_mm_q4_1_f32);
GGML_METAL_ADD_KERNEL(mul_mm_q2_K_f32);
GGML_METAL_ADD_KERNEL(mul_mm_q3_K_f32);
GGML_METAL_ADD_KERNEL(mul_mm_q4_K_f32);
GGML_METAL_ADD_KERNEL(mul_mm_q5_K_f32);
GGML_METAL_ADD_KERNEL(mul_mm_q6_K_f32);
GGML_METAL_ADD_KERNEL(mul_mv_f32_f32);
GGML_METAL_ADD_KERNEL(mul_mv_f16_f32);
GGML_METAL_ADD_KERNEL(mul_mv_f16_f32_1row);
GGML_METAL_ADD_KERNEL(mul_mv_f16_f32_l4);
GGML_METAL_ADD_KERNEL(mul_mv_q4_0_f32);
GGML_METAL_ADD_KERNEL(mul_mv_q4_1_f32);
GGML_METAL_ADD_KERNEL(mul_mv_q8_0_f32);
GGML_METAL_ADD_KERNEL(mul_mv_q2_K_f32);
GGML_METAL_ADD_KERNEL(mul_mv_q3_K_f32);
GGML_METAL_ADD_KERNEL(mul_mv_q4_K_f32);
GGML_METAL_ADD_KERNEL(mul_mv_q5_K_f32);
GGML_METAL_ADD_KERNEL(mul_mv_q6_K_f32);
if ([ctx->device supportsFamily:MTLGPUFamilyApple7]) {
GGML_METAL_ADD_KERNEL(mul_mm_f32_f32);
GGML_METAL_ADD_KERNEL(mul_mm_f16_f32);
GGML_METAL_ADD_KERNEL(mul_mm_q4_0_f32);
GGML_METAL_ADD_KERNEL(mul_mm_q8_0_f32);
GGML_METAL_ADD_KERNEL(mul_mm_q4_1_f32);
GGML_METAL_ADD_KERNEL(mul_mm_q2_K_f32);
GGML_METAL_ADD_KERNEL(mul_mm_q3_K_f32);
GGML_METAL_ADD_KERNEL(mul_mm_q4_K_f32);
GGML_METAL_ADD_KERNEL(mul_mm_q5_K_f32);
GGML_METAL_ADD_KERNEL(mul_mm_q6_K_f32);
}
GGML_METAL_ADD_KERNEL(rope_f32);
GGML_METAL_ADD_KERNEL(rope_f16);
GGML_METAL_ADD_KERNEL(alibi_f32);
@ -308,8 +298,21 @@ struct ggml_metal_context * ggml_metal_init(int n_cb) {
#undef GGML_METAL_ADD_KERNEL
}
GGML_METAL_LOG_INFO("%s: hasUnifiedMemory = %s\n", __func__, ctx->device.hasUnifiedMemory ? "true" : "false");
#if TARGET_OS_OSX
// print MTL GPU family:
GGML_METAL_LOG_INFO("%s: GPU name: %s\n", __func__, [[ctx->device name] UTF8String]);
// determine max supported GPU family
// https://developer.apple.com/metal/Metal-Shading-Language-Specification.pdf
// https://developer.apple.com/metal/Metal-Feature-Set-Tables.pdf
for (int i = MTLGPUFamilyApple1 + 20; i >= MTLGPUFamilyApple1; --i) {
if ([ctx->device supportsFamily:i]) {
GGML_METAL_LOG_INFO("%s: GPU family: MTLGPUFamilyApple%d (%d)\n", __func__, i - MTLGPUFamilyApple1 + 1, i);
break;
}
}
GGML_METAL_LOG_INFO("%s: hasUnifiedMemory = %s\n", __func__, ctx->device.hasUnifiedMemory ? "true" : "false");
GGML_METAL_LOG_INFO("%s: recommendedMaxWorkingSetSize = %8.2f MB\n", __func__, ctx->device.recommendedMaxWorkingSetSize / 1024.0 / 1024.0);
if (ctx->device.maxTransferRate != 0) {
GGML_METAL_LOG_INFO("%s: maxTransferRate = %8.2f MB/s\n", __func__, ctx->device.maxTransferRate / 1024.0 / 1024.0);
@ -351,28 +354,30 @@ void ggml_metal_free(struct ggml_metal_context * ctx) {
GGML_METAL_DEL_KERNEL(get_rows_q6_K);
GGML_METAL_DEL_KERNEL(rms_norm);
GGML_METAL_DEL_KERNEL(norm);
GGML_METAL_DEL_KERNEL(mul_mat_f32_f32);
GGML_METAL_DEL_KERNEL(mul_mat_f16_f32);
GGML_METAL_DEL_KERNEL(mul_mat_f16_f32_1row);
GGML_METAL_DEL_KERNEL(mul_mat_f16_f32_l4);
GGML_METAL_DEL_KERNEL(mul_mat_q4_0_f32);
GGML_METAL_DEL_KERNEL(mul_mat_q4_1_f32);
GGML_METAL_DEL_KERNEL(mul_mat_q8_0_f32);
GGML_METAL_DEL_KERNEL(mul_mat_q2_K_f32);
GGML_METAL_DEL_KERNEL(mul_mat_q3_K_f32);
GGML_METAL_DEL_KERNEL(mul_mat_q4_K_f32);
GGML_METAL_DEL_KERNEL(mul_mat_q5_K_f32);
GGML_METAL_DEL_KERNEL(mul_mat_q6_K_f32);
GGML_METAL_DEL_KERNEL(mul_mm_f32_f32);
GGML_METAL_DEL_KERNEL(mul_mm_f16_f32);
GGML_METAL_DEL_KERNEL(mul_mm_q4_0_f32);
GGML_METAL_DEL_KERNEL(mul_mm_q8_0_f32);
GGML_METAL_DEL_KERNEL(mul_mm_q4_1_f32);
GGML_METAL_DEL_KERNEL(mul_mm_q2_K_f32);
GGML_METAL_DEL_KERNEL(mul_mm_q3_K_f32);
GGML_METAL_DEL_KERNEL(mul_mm_q4_K_f32);
GGML_METAL_DEL_KERNEL(mul_mm_q5_K_f32);
GGML_METAL_DEL_KERNEL(mul_mm_q6_K_f32);
GGML_METAL_DEL_KERNEL(mul_mv_f32_f32);
GGML_METAL_DEL_KERNEL(mul_mv_f16_f32);
GGML_METAL_DEL_KERNEL(mul_mv_f16_f32_1row);
GGML_METAL_DEL_KERNEL(mul_mv_f16_f32_l4);
GGML_METAL_DEL_KERNEL(mul_mv_q4_0_f32);
GGML_METAL_DEL_KERNEL(mul_mv_q4_1_f32);
GGML_METAL_DEL_KERNEL(mul_mv_q8_0_f32);
GGML_METAL_DEL_KERNEL(mul_mv_q2_K_f32);
GGML_METAL_DEL_KERNEL(mul_mv_q3_K_f32);
GGML_METAL_DEL_KERNEL(mul_mv_q4_K_f32);
GGML_METAL_DEL_KERNEL(mul_mv_q5_K_f32);
GGML_METAL_DEL_KERNEL(mul_mv_q6_K_f32);
if ([ctx->device supportsFamily:MTLGPUFamilyApple7]) {
GGML_METAL_DEL_KERNEL(mul_mm_f32_f32);
GGML_METAL_DEL_KERNEL(mul_mm_f16_f32);
GGML_METAL_DEL_KERNEL(mul_mm_q4_0_f32);
GGML_METAL_DEL_KERNEL(mul_mm_q8_0_f32);
GGML_METAL_DEL_KERNEL(mul_mm_q4_1_f32);
GGML_METAL_DEL_KERNEL(mul_mm_q2_K_f32);
GGML_METAL_DEL_KERNEL(mul_mm_q3_K_f32);
GGML_METAL_DEL_KERNEL(mul_mm_q4_K_f32);
GGML_METAL_DEL_KERNEL(mul_mm_q5_K_f32);
GGML_METAL_DEL_KERNEL(mul_mm_q6_K_f32);
}
GGML_METAL_DEL_KERNEL(rope_f32);
GGML_METAL_DEL_KERNEL(rope_f16);
GGML_METAL_DEL_KERNEL(alibi_f32);
@ -437,7 +442,7 @@ static id<MTLBuffer> ggml_metal_get_buffer(struct ggml_metal_context * ctx, stru
for (int i = 0; i < ctx->n_buffers; ++i) {
const int64_t ioffs = (int64_t) t->data - (int64_t) ctx->buffers[i].data;
//metal_printf("ioffs = %10ld, tsize = %10ld, sum = %10ld, ctx->buffers[%d].size = %10ld, name = %s\n", ioffs, tsize, ioffs + tsize, i, ctx->buffers[i].size, ctx->buffers[i].name);
//GGML_METAL_LOG_INFO("ioffs = %10ld, tsize = %10ld, sum = %10ld, ctx->buffers[%d].size = %10ld, name = %s\n", ioffs, tsize, ioffs + tsize, i, ctx->buffers[i].size, ctx->buffers[i].name);
if (ioffs >= 0 && ioffs + tsize <= (int64_t) ctx->buffers[i].size) {
*offs = (size_t) ioffs;
@ -1002,21 +1007,46 @@ void ggml_metal_graph_compute(
} break;
case GGML_OP_MUL_MAT:
{
// TODO: needs to be updated after PR: https://github.com/ggerganov/ggml/pull/224
GGML_ASSERT(ne00 == ne10);
// GGML_ASSERT(ne02 == ne12); // Should be checked on individual data types until broadcast is implemented everywhere
uint gqa = ne12/ne02;
GGML_ASSERT(ne03 == ne13);
const uint gqa = ne12/ne02;
// find the break-even point where the matrix-matrix kernel becomes more efficient compared
// to the matrix-vector kernel
int ne11_mm_min = 1;
#if 0
// the numbers below are measured on M2 Ultra for 7B and 13B models
// these numbers do not translate to other devices or model sizes
// TODO: need to find a better approach
if ([ctx->device.name isEqualToString:@"Apple M2 Ultra"]) {
switch (src0t) {
case GGML_TYPE_F16: ne11_mm_min = 2; break;
case GGML_TYPE_Q8_0: ne11_mm_min = 7; break;
case GGML_TYPE_Q2_K: ne11_mm_min = 15; break;
case GGML_TYPE_Q3_K: ne11_mm_min = 7; break;
case GGML_TYPE_Q4_0:
case GGML_TYPE_Q4_1: ne11_mm_min = 15; break;
case GGML_TYPE_Q4_K: ne11_mm_min = 11; break;
case GGML_TYPE_Q5_0: // not tested yet
case GGML_TYPE_Q5_1: ne11_mm_min = 13; break; // not tested yet
case GGML_TYPE_Q5_K: ne11_mm_min = 7; break;
case GGML_TYPE_Q6_K: ne11_mm_min = 7; break;
default: ne11_mm_min = 1; break;
}
}
#endif
// for now the matrix-matrix multiplication kernel only works on A14+/M1+ SoCs
// AMD GPU and older A-chips will reuse matrix-vector multiplication kernel
if (!ggml_is_transposed(src0) &&
if ([ctx->device supportsFamily:MTLGPUFamilyApple7] &&
!ggml_is_transposed(src0) &&
!ggml_is_transposed(src1) &&
src1t == GGML_TYPE_F32 &&
[ctx->device supportsFamily:MTLGPUFamilyApple7] &&
ne00%32 == 0 &&
ne11 > 2) {
ne00 % 32 == 0 &&
ne11 > ne11_mm_min) {
//printf("matrix: ne00 = %6d, ne01 = %6d, ne02 = %6d, ne11 = %6d, ne12 = %6d\n", ne00, ne01, ne02, ne11, ne12);
switch (src0->type) {
case GGML_TYPE_F32: [encoder setComputePipelineState:ctx->pipeline_mul_mm_f32_f32]; break;
case GGML_TYPE_F16: [encoder setComputePipelineState:ctx->pipeline_mul_mm_f16_f32]; break;
@ -1045,17 +1075,18 @@ void ggml_metal_graph_compute(
[encoder setBytes:&ne1 length:sizeof(ne1) atIndex:12];
[encoder setBytes:&gqa length:sizeof(gqa) atIndex:13];
[encoder setThreadgroupMemoryLength:8192 atIndex:0];
[encoder dispatchThreadgroups:MTLSizeMake( (ne11+31)/32, (ne01+63) / 64, ne12) threadsPerThreadgroup:MTLSizeMake(128, 1, 1)];
[encoder dispatchThreadgroups:MTLSizeMake( (ne11 + 31)/32, (ne01 + 63)/64, ne12) threadsPerThreadgroup:MTLSizeMake(128, 1, 1)];
} else {
int nth0 = 32;
int nth1 = 1;
int nrows = 1;
//printf("vector: ne00 = %6d, ne01 = %6d, ne02 = %6d, ne11 = %6d, ne12 = %6d\n", ne00, ne01, ne02, ne11, ne12);
// use custom matrix x vector kernel
switch (src0t) {
case GGML_TYPE_F32:
{
[encoder setComputePipelineState:ctx->pipeline_mul_mat_f32_f32];
[encoder setComputePipelineState:ctx->pipeline_mul_mv_f32_f32];
nrows = 4;
} break;
case GGML_TYPE_F16:
@ -1063,12 +1094,12 @@ void ggml_metal_graph_compute(
nth0 = 32;
nth1 = 1;
if (ne11 * ne12 < 4) {
[encoder setComputePipelineState:ctx->pipeline_mul_mat_f16_f32_1row];
[encoder setComputePipelineState:ctx->pipeline_mul_mv_f16_f32_1row];
} else if (ne00 >= 128 && ne01 >= 8 && ne00%4 == 0) {
[encoder setComputePipelineState:ctx->pipeline_mul_mat_f16_f32_l4];
[encoder setComputePipelineState:ctx->pipeline_mul_mv_f16_f32_l4];
nrows = ne11;
} else {
[encoder setComputePipelineState:ctx->pipeline_mul_mat_f16_f32];
[encoder setComputePipelineState:ctx->pipeline_mul_mv_f16_f32];
nrows = 4;
}
} break;
@ -1079,7 +1110,7 @@ void ggml_metal_graph_compute(
nth0 = 8;
nth1 = 8;
[encoder setComputePipelineState:ctx->pipeline_mul_mat_q4_0_f32];
[encoder setComputePipelineState:ctx->pipeline_mul_mv_q4_0_f32];
} break;
case GGML_TYPE_Q4_1:
{
@ -1088,7 +1119,7 @@ void ggml_metal_graph_compute(
nth0 = 8;
nth1 = 8;
[encoder setComputePipelineState:ctx->pipeline_mul_mat_q4_1_f32];
[encoder setComputePipelineState:ctx->pipeline_mul_mv_q4_1_f32];
} break;
case GGML_TYPE_Q8_0:
{
@ -1097,7 +1128,7 @@ void ggml_metal_graph_compute(
nth0 = 8;
nth1 = 8;
[encoder setComputePipelineState:ctx->pipeline_mul_mat_q8_0_f32];
[encoder setComputePipelineState:ctx->pipeline_mul_mv_q8_0_f32];
} break;
case GGML_TYPE_Q2_K:
{
@ -1106,7 +1137,7 @@ void ggml_metal_graph_compute(
nth0 = 2;
nth1 = 32;
[encoder setComputePipelineState:ctx->pipeline_mul_mat_q2_K_f32];
[encoder setComputePipelineState:ctx->pipeline_mul_mv_q2_K_f32];
} break;
case GGML_TYPE_Q3_K:
{
@ -1115,7 +1146,7 @@ void ggml_metal_graph_compute(
nth0 = 2;
nth1 = 32;
[encoder setComputePipelineState:ctx->pipeline_mul_mat_q3_K_f32];
[encoder setComputePipelineState:ctx->pipeline_mul_mv_q3_K_f32];
} break;
case GGML_TYPE_Q4_K:
{
@ -1124,7 +1155,7 @@ void ggml_metal_graph_compute(
nth0 = 4; //1;
nth1 = 8; //32;
[encoder setComputePipelineState:ctx->pipeline_mul_mat_q4_K_f32];
[encoder setComputePipelineState:ctx->pipeline_mul_mv_q4_K_f32];
} break;
case GGML_TYPE_Q5_K:
{
@ -1133,7 +1164,7 @@ void ggml_metal_graph_compute(
nth0 = 2;
nth1 = 32;
[encoder setComputePipelineState:ctx->pipeline_mul_mat_q5_K_f32];
[encoder setComputePipelineState:ctx->pipeline_mul_mv_q5_K_f32];
} break;
case GGML_TYPE_Q6_K:
{
@ -1142,7 +1173,7 @@ void ggml_metal_graph_compute(
nth0 = 2;
nth1 = 32;
[encoder setComputePipelineState:ctx->pipeline_mul_mat_q6_K_f32];
[encoder setComputePipelineState:ctx->pipeline_mul_mv_q6_K_f32];
} break;
default:
{
@ -1171,7 +1202,7 @@ void ggml_metal_graph_compute(
[encoder setBytes:&gqa length:sizeof(gqa) atIndex:17];
if (src0t == GGML_TYPE_Q4_0 || src0t == GGML_TYPE_Q4_1 || src0t == GGML_TYPE_Q8_0 ||
src0t == GGML_TYPE_Q2_K) {// || src0t == GGML_TYPE_Q4_K) {
src0t == GGML_TYPE_Q2_K) { // || src0t == GGML_TYPE_Q4_K) {
[encoder dispatchThreadgroups:MTLSizeMake((ne01 + 7)/8, ne11, ne12) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
}
else if (src0t == GGML_TYPE_Q4_K) {