vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-11 01:24:36 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merge branch 'master' into concedo_experimental

Browse source 
# Conflicts:
#	.ecrc
#	.github/workflows/build.yml
#	CMakeLists.txt
#	README.md
#	llama.cpp
#	tests/test-c.c
This commit is contained in:
Concedo 2024-01-30 19:18:10 +08:00
commit 54cc31f9dc
48 changed files with 4272 additions and 73 deletions
Download patch file Download diff file Expand all files Collapse all files

43
llama.cpp
View file

@ -17,9 +17,11 @@
#endif
#if defined(GGML_USE_VULKAN)
# include "ggml-vulkan.h"
#elif defined(GGML_USE_SYCL)
# include "ggml-sycl.h"
#endif
#if defined(GGML_USE_SYCL)
#include "ggml-sycl.h"
#if defined(GGML_USE_KOMPUTE)
# include "ggml-kompute.h"
#endif
#ifdef GGML_USE_METAL
@ -1186,10 +1188,10 @@ struct llama_mlock {
#ifdef __APPLE__
#define MLOCK_SUGGESTION \
"Try increasing the sysctl values 'vm.user_wire_limit' and 'vm.global_user_wire_limit' and/or " \
"decreasing 'vm.global_no_user_wire_amount'. Also try increasing RLIMIT_MLOCK (ulimit -l).\n"
"decreasing 'vm.global_no_user_wire_amount'. Also try increasing RLIMIT_MEMLOCK (ulimit -l).\n"
#else
#define MLOCK_SUGGESTION \
"Try increasing RLIMIT_MLOCK ('ulimit -l' as root).\n"
"Try increasing RLIMIT_MEMLOCK ('ulimit -l' as root).\n"
#endif
bool raw_lock(const void * addr, size_t size) const {
@ -1341,6 +1343,11 @@ static ggml_backend_buffer_type_t llama_default_buffer_type_offload(int gpu) {
buft = ggml_backend_sycl_buffer_type(gpu);
#elif defined(GGML_USE_CLBLAST)
buft = ggml_backend_opencl_buffer_type();
#elif defined(GGML_USE_KOMPUTE)
buft = ggml_backend_kompute_buffer_type(gpu);
if (buft == nullptr) {
LLAMA_LOG_WARN("%s: cannot use GPU %d, check `vulkaninfo --summary`\n", __func__, gpu);
}
#endif
if (buft == nullptr) {
@ -4180,7 +4187,7 @@ static bool llm_load_tensors(
}
// Returns 0 on success, -1 on error, and -2 on cancellation via llama_progress_callback
static int llama_model_load(const std::string & fname, llama_model & model, const llama_model_params & params) {
static int llama_model_load(const std::string & fname, llama_model & model, llama_model_params & params) {
try {
llama_model_loader ml(fname, params.use_mmap, params.kv_overrides);
@ -4201,6 +4208,22 @@ static int llama_model_load(const std::string & fname, llama_model & model, cons
return 0;
}
#ifdef GGML_USE_KOMPUTE
if (params.n_gpu_layers > 0 && (
!(model.arch == LLM_ARCH_LLAMA || model.arch == LLM_ARCH_FALCON)
|| !(
model.ftype == LLAMA_FTYPE_ALL_F32 ||
model.ftype == LLAMA_FTYPE_MOSTLY_F16 ||
model.ftype == LLAMA_FTYPE_MOSTLY_Q4_0 ||
model.ftype == LLAMA_FTYPE_MOSTLY_Q4_1
)
)) {
// TODO(cebtenzzre): propagate this error outside of llama_load_model_from_file
LLAMA_LOG_WARN("%s: disabling Kompute due to unsupported model arch or quantization\n", __func__);
params.n_gpu_layers = 0;
}
#endif
if (!llm_load_tensors(
ml, model, params.n_gpu_layers, params.split_mode, params.main_gpu, params.tensor_split, params.use_mlock,
params.progress_callback, params.progress_callback_user_data
@ -10571,6 +10594,16 @@ struct llama_context * llama_new_context_with_model(
}
ctx->backends.push_back(backend);
}
#elif defined(GGML_USE_KOMPUTE)
if (model->n_gpu_layers > 0) {
auto * backend = ggml_backend_kompute_init(model->main_gpu);
if (backend == nullptr) {
LLAMA_LOG_ERROR("%s: failed to initialize Kompute backend\n", __func__);
llama_free(ctx);
return nullptr;
}
ctx->backends.push_back(backend);
}
#endif
ctx->backend_cpu = ggml_backend_cpu_init();
if (ctx->backend_cpu == nullptr) {