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

Merge remote-tracking branch 'origin/master' into concedo_experimental

Browse source 
# Conflicts:
#	README.md
This commit is contained in:
Concedo 2023-06-18 22:53:10 +08:00
commit d0d3c4f32b
3 changed files with 21 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

6
examples/metal/metal.cpp
View file

@ -40,8 +40,10 @@ int main(int argc, char ** argv) {
// this allocates all Metal resources and memory buffers // this allocates all Metal resources and memory buffers
auto * ctx_metal = ggml_metal_init(); auto * ctx_metal = ggml_metal_init();
ggml_metal_add_buffer(ctx_metal, "data", ggml_get_mem_buffer(ctx_data), ggml_get_mem_size(ctx_data)); const size_t max_size_data = ggml_get_max_tensor_size(ctx_data);
ggml_metal_add_buffer(ctx_metal, "eval", ggml_get_mem_buffer(ctx_eval), ggml_get_mem_size(ctx_eval)); const size_t max_size_eval = ggml_get_max_tensor_size(ctx_eval);
ggml_metal_add_buffer(ctx_metal, "data", ggml_get_mem_buffer(ctx_data), ggml_get_mem_size(ctx_data), max_size_data);
ggml_metal_add_buffer(ctx_metal, "eval", ggml_get_mem_buffer(ctx_eval), ggml_get_mem_size(ctx_eval), max_size_eval);
// main // main
{ {

2
koboldcpp.py
View file

@ -225,7 +225,7 @@ maxhordectx = 1024
maxhordelen = 256 maxhordelen = 256
modelbusy = False modelbusy = False
defaultport = 5001 defaultport = 5001
KcppVersion = "1.31.2" KcppVersion = "1.32"
showdebug = True showdebug = True
class ServerRequestHandler(http.server.SimpleHTTPRequestHandler): class ServerRequestHandler(http.server.SimpleHTTPRequestHandler):

16
llama.cpp
View file

@ -19,6 +19,11 @@
#ifdef GGML_USE_METAL #ifdef GGML_USE_METAL
#include "ggml-metal.h" #include "ggml-metal.h"
#endif #endif
#ifdef GGML_USE_K_QUANTS
#ifndef QK_K
#define QK_K 256
#endif
#endif
#include <array> #include <array>
#include <ctime> #include <ctime>
@ -2491,6 +2496,17 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
} else { } else {
new_type = quantized_type; new_type = quantized_type;
#ifdef GGML_USE_K_QUANTS #ifdef GGML_USE_K_QUANTS
if (quantized_type == GGML_TYPE_Q2_K || quantized_type == GGML_TYPE_Q3_K || quantized_type == GGML_TYPE_Q4_K ||
quantized_type == GGML_TYPE_Q5_K || quantized_type == GGML_TYPE_Q6_K) {
int nx = tensor.ne.at(0);
int ny = tensor.ne.at(0);
if (nx % QK_K != 0 || ny % QK_K != 0) {
fprintf(stderr, "\n\n========================= Tensor sizes %d x %d are not divisible by %d\n",nx,ny,QK_K);
fprintf(stderr, "This is required to be able to use k-quants for now!\n");
fprintf(stderr, "========================================================================================\n\n");
throw std::runtime_error("Unsupported tensor size encountered\n");
}
}
if (tensor.name == "output.weight") { if (tensor.name == "output.weight") {
new_type = GGML_TYPE_Q6_K; new_type = GGML_TYPE_Q6_K;
} else if (tensor.name.find("attention.wv.weight") != std::string::npos) { } else if (tensor.name.find("attention.wv.weight") != std::string::npos) {