Merge branch 'master' of https://github.com/ggerganov/llama.cpp into clang-warnings
This commit is contained in:
Cebtenzzre
commit
d38d59cc91
11 changed files with 33 additions and 23 deletions
1
.github/workflows/build.yml
vendored
1
.github/workflows/build.yml
vendored
|
|
@ -468,6 +468,7 @@ jobs:
|
|||
with:
|
||||
operating_system: freebsd
|
||||
version: '13.2'
|
||||
hypervisor: 'qemu'
|
||||
run: |
|
||||
sudo pkg update
|
||||
sudo pkg install -y gmake automake autoconf pkgconf llvm15 clinfo clover opencl clblast openblas
|
||||
|
|
|
|||
14
Makefile
14
Makefile
|
|
@ -542,22 +542,22 @@ main: examples/main/main.cpp build-info.h ggml.
|
|||
@echo '==== Run ./main -h for help. ===='
|
||||
@echo
|
||||
|
||||
simple: examples/simple/simple.cpp ggml.o llama.o common.o $(OBJS)
|
||||
simple: examples/simple/simple.cpp build-info.h ggml.o llama.o common.o $(OBJS)
|
||||
$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
|
||||
|
||||
quantize: examples/quantize/quantize.cpp ggml.o llama.o $(OBJS)
|
||||
quantize: examples/quantize/quantize.cpp build-info.h ggml.o llama.o $(OBJS)
|
||||
$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
|
||||
|
||||
quantize-stats: examples/quantize-stats/quantize-stats.cpp ggml.o llama.o $(OBJS)
|
||||
quantize-stats: examples/quantize-stats/quantize-stats.cpp build-info.h ggml.o llama.o $(OBJS)
|
||||
$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
|
||||
|
||||
perplexity: examples/perplexity/perplexity.cpp ggml.o llama.o common.o $(OBJS)
|
||||
perplexity: examples/perplexity/perplexity.cpp build-info.h ggml.o llama.o common.o $(OBJS)
|
||||
$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
|
||||
|
||||
embedding: examples/embedding/embedding.cpp ggml.o llama.o common.o $(OBJS)
|
||||
embedding: examples/embedding/embedding.cpp build-info.h ggml.o llama.o common.o $(OBJS)
|
||||
$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
|
||||
|
||||
save-load-state: examples/save-load-state/save-load-state.cpp ggml.o llama.o common.o $(OBJS)
|
||||
save-load-state: examples/save-load-state/save-load-state.cpp build-info.h ggml.o llama.o common.o $(OBJS)
|
||||
$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
|
||||
|
||||
server: examples/server/server.cpp examples/server/httplib.h examples/server/json.hpp examples/server/index.html.hpp examples/server/index.js.hpp examples/server/completion.js.hpp build-info.h ggml.o llama.o common.o grammar-parser.o $(OBJS)
|
||||
|
|
@ -610,7 +610,7 @@ build-info.h: $(wildcard .git/index) scripts/build-info.sh
|
|||
|
||||
tests: $(TEST_TARGETS)
|
||||
|
||||
benchmark-matmult: examples/benchmark/benchmark-matmult.cpp ggml.o $(OBJS)
|
||||
benchmark-matmult: examples/benchmark/benchmark-matmult.cpp build-info.h ggml.o $(OBJS)
|
||||
$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
|
||||
|
||||
run-benchmark-matmult: benchmark-matmult
|
||||
|
|
|
|||
|
|
@ -3,7 +3,6 @@
|
|||
#pragma once
|
||||
|
||||
#include "llama.h"
|
||||
#include "build-info.h"
|
||||
|
||||
#define LOG_NO_FILE_LINE_FUNCTION
|
||||
#include "log.h"
|
||||
|
|
|
|||
|
|
@ -1,3 +1,4 @@
|
|||
#include "build-info.h"
|
||||
#include "common.h"
|
||||
#include "ggml.h"
|
||||
|
||||
|
|
@ -32,11 +33,11 @@ static void ggml_graph_compute_helper(std::vector<uint8_t> & buf, ggml_cgraph *
|
|||
}
|
||||
|
||||
static float tensor_sum_elements(const ggml_tensor * tensor) {
|
||||
float sum = 0;
|
||||
if (tensor->type==GGML_TYPE_F32) {
|
||||
double sum = 0;
|
||||
if (tensor->type == GGML_TYPE_F32) {
|
||||
for (int j = 0; j < tensor->ne[1]; j++) {
|
||||
for (int k = 0; k < tensor->ne[0]; k++) {
|
||||
sum += ((float *) tensor->data)[j*tensor->ne[0]+k];
|
||||
sum += ((float *) tensor->data)[j*tensor->ne[0] + k];
|
||||
}
|
||||
}
|
||||
}
|
||||
|
|
@ -125,12 +126,15 @@ int main(int argc, char ** argv) {
|
|||
|
||||
//printf("Memsize required = %i\n", sizex*sizex);
|
||||
|
||||
// TODO: perform the bench for all types or for a user specified type
|
||||
const ggml_type qtype = GGML_TYPE_Q4_1;
|
||||
|
||||
size_t ctx_size = 0;
|
||||
ctx_size += sizex*sizey*ggml_type_sizef(GGML_TYPE_F32);
|
||||
ctx_size += sizex*sizey*ggml_type_sizef(GGML_TYPE_F32);
|
||||
ctx_size += sizex*sizez*ggml_type_sizef(GGML_TYPE_F32);
|
||||
ctx_size += sizex*sizey*ggml_type_sizef(GGML_TYPE_Q4_0);
|
||||
ctx_size += sizex*sizey*ggml_type_sizef(GGML_TYPE_Q4_0);
|
||||
ctx_size += sizex*sizey*ggml_type_sizef(qtype);
|
||||
ctx_size += sizex*sizey*ggml_type_sizef(qtype);
|
||||
ctx_size += sizex*sizey*ggml_type_sizef(GGML_TYPE_F32); // BLAS
|
||||
ctx_size += sizex*sizey*ggml_type_sizef(GGML_TYPE_F32); // BLAS
|
||||
ctx_size += 1024*1024*16;
|
||||
|
|
@ -163,7 +167,7 @@ int main(int argc, char ** argv) {
|
|||
struct ggml_tensor * m2 = ggml_new_tensor_2d(ctx, GGML_TYPE_F32, sizex, sizez);
|
||||
ggml_set_f32(m2, 2.0f);
|
||||
|
||||
printf("\n------ Test 1 - Matrix Mult via F32 code ------------------------------------------------------------------------------\n");
|
||||
printf("\n------ Test 1 - Matrix Mult via F32 code\n");
|
||||
// printf("Creating new tensor m11xm2\n");
|
||||
struct ggml_tensor * m11xm2 = ggml_mul_mat(ctx, m11, m2);
|
||||
|
||||
|
|
@ -181,17 +185,16 @@ int main(int argc, char ** argv) {
|
|||
|
||||
TENSOR_DUMP(gf.nodes[0]);
|
||||
|
||||
printf("\n------ Test 2 - Matrix Mult via Q4_0 code ------------------------------------------------------------------------------\n");
|
||||
printf("\n------ Test 2 - Matrix Mult via %s code\n", ggml_type_name(qtype));
|
||||
|
||||
int32_t nelements = sizex*sizey;
|
||||
int32_t ne[2] = { sizex, sizey };
|
||||
|
||||
std::vector<int64_t> hist_cur(1 << 4, 0);
|
||||
|
||||
// Set up a the benchmark matrices
|
||||
// printf("Creating new tensor q11 & Running quantize\n");
|
||||
struct ggml_tensor * q11 = ggml_new_tensor_2d(ctx, GGML_TYPE_Q4_0, sizex, sizey);
|
||||
ggml_quantize_q4_0((const float *) m11->data, q11->data, nelements, ne[0], hist_cur.data());
|
||||
struct ggml_tensor * q11 = ggml_new_tensor_2d(ctx, qtype, sizex, sizey);
|
||||
ggml_quantize_chunk(qtype, (const float *) m11->data, q11->data, 0, nelements, hist_cur.data());
|
||||
|
||||
// Set up a the compute graph
|
||||
// printf("Creating new tensor q31\n");
|
||||
|
|
@ -202,8 +205,8 @@ int main(int argc, char ** argv) {
|
|||
|
||||
// Set up a second graph computation to make sure we override the CPU cache lines
|
||||
// printf("Creating new tensor q12 & Running quantize\n");
|
||||
struct ggml_tensor * q12 = ggml_new_tensor_2d(ctx, GGML_TYPE_Q4_0, sizex, sizey);
|
||||
ggml_quantize_q4_0((const float *) m12->data, q12->data, nelements, ne[0], hist_cur.data());
|
||||
struct ggml_tensor * q12 = ggml_new_tensor_2d(ctx, qtype, sizex, sizey);
|
||||
ggml_quantize_chunk(qtype, (const float *) m12->data, q12->data, 0, nelements, hist_cur.data());
|
||||
|
||||
// printf("Creating new tensor q32\n");
|
||||
struct ggml_tensor * q32 = ggml_mul_mat(ctx, q12, m2);
|
||||
|
|
@ -220,7 +223,7 @@ int main(int argc, char ** argv) {
|
|||
printf("Matrix Multiplication of (%i,%i,%i) x (%i,%i,%i) - about %6.2f gFLOPS\n\n", sizex, sizey, 1, sizex, sizez, 1, 1.0f*flops_per_matrix / 1000 / 1000 / 1000);
|
||||
|
||||
|
||||
// Let's use the F32 result from above as a reference for the q4_0 multiplication
|
||||
// Let's use the F32 result from above as a reference for the quantized multiplication
|
||||
float sum_of_F32_reference = tensor_sum_elements(gf.nodes[0]);
|
||||
|
||||
printf("Iteration;NThreads; SizeX; SizeY; SizeZ; Required_FLOPS; Elapsed_u_Seconds; gigaFLOPS\n");
|
||||
|
|
@ -250,7 +253,7 @@ int main(int argc, char ** argv) {
|
|||
// Check that the matrix multiplication result is in the right ballpark
|
||||
// We cannot use the exact value from the F32 multiplication because the quantizuation will be slightly different
|
||||
float sum_of_Q4_result = tensor_sum_elements(gf31.nodes[0]);
|
||||
float delta = abs(sum_of_Q4_result - sum_of_F32_reference);
|
||||
float delta = std::abs(sum_of_Q4_result - sum_of_F32_reference);
|
||||
float allowed_delta = (sum_of_F32_reference) / 1000 / 1000; // Let's accept an epsilon of 10^-6
|
||||
|
||||
if (delta > allowed_delta) {
|
||||
|
|
|
|||
|
|
@ -1,3 +1,4 @@
|
|||
#include "build-info.h"
|
||||
#include "common.h"
|
||||
#include "embd-input.h"
|
||||
|
||||
|
|
|
|||
|
|
@ -1,3 +1,4 @@
|
|||
#include "build-info.h"
|
||||
#include "common.h"
|
||||
#include "llama.h"
|
||||
|
||||
|
|
|
|||
|
|
@ -1,3 +1,4 @@
|
|||
#include "build-info.h"
|
||||
#include "common.h"
|
||||
#include "llama.h"
|
||||
|
||||
|
|
|
|||
|
|
@ -1,4 +1,5 @@
|
|||
#define LLAMA_API_INTERNAL
|
||||
#include "build-info.h"
|
||||
#include "common.h"
|
||||
#include "ggml.h"
|
||||
#include "llama.h"
|
||||
|
|
|
|||
|
|
@ -1,3 +1,4 @@
|
|||
#include "build-info.h"
|
||||
#include "common.h"
|
||||
#include "llama.h"
|
||||
|
||||
|
|
|
|||
|
|
@ -1,3 +1,4 @@
|
|||
#include "build-info.h"
|
||||
#include "common.h"
|
||||
#include "llama.h"
|
||||
|
||||
|
|
|
|||
|
|
@ -52,7 +52,8 @@
|
|||
in
|
||||
{
|
||||
packages.default = pkgs.stdenv.mkDerivation {
|
||||
inherit name src meta postPatch nativeBuildInputs buildInputs postInstall;
|
||||
inherit name src meta postPatch nativeBuildInputs postInstall;
|
||||
buildInputs = osSpecific;
|
||||
cmakeFlags = cmakeFlags
|
||||
++ (if isAarch64 && isDarwin then [
|
||||
"-DCMAKE_C_FLAGS=-D__ARM_FEATURE_DOTPROD=1"
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue