mirror of
https://github.com/nomic-ai/gpt4all.git
synced 2025-08-13 05:35:45 +00:00
Link against ggml in bin so we can get the available devices without loading a model.
This commit is contained in:
parent
0f046cf905
commit
045f6e6cdc
@ -134,6 +134,8 @@ add_library(llmodel
|
|||||||
llmodel_c.h llmodel_c.cpp
|
llmodel_c.h llmodel_c.cpp
|
||||||
dlhandle.h
|
dlhandle.h
|
||||||
)
|
)
|
||||||
|
target_link_libraries(llmodel PRIVATE ggml-mainline-default)
|
||||||
|
target_compile_definitions(llmodel PRIVATE GGML_BUILD_VARIANT="default")
|
||||||
target_compile_definitions(llmodel PRIVATE LIB_FILE_EXT="${CMAKE_SHARED_LIBRARY_SUFFIX}")
|
target_compile_definitions(llmodel PRIVATE LIB_FILE_EXT="${CMAKE_SHARED_LIBRARY_SUFFIX}")
|
||||||
|
|
||||||
set_target_properties(llmodel PROPERTIES
|
set_target_properties(llmodel PROPERTIES
|
||||||
|
@ -101,6 +101,7 @@ public:
|
|||||||
virtual bool initializeGPUDevice(int /*device*/) { return false; }
|
virtual bool initializeGPUDevice(int /*device*/) { return false; }
|
||||||
virtual bool hasGPUDevice() { return false; }
|
virtual bool hasGPUDevice() { return false; }
|
||||||
virtual bool usingGPUDevice() { return false; }
|
virtual bool usingGPUDevice() { return false; }
|
||||||
|
static std::vector<GPUDevice> availableGPUDevices();
|
||||||
|
|
||||||
protected:
|
protected:
|
||||||
// These are pure virtual because subclasses need to implement as the default implementation of
|
// These are pure virtual because subclasses need to implement as the default implementation of
|
||||||
|
@ -4,6 +4,10 @@
|
|||||||
#include <iostream>
|
#include <iostream>
|
||||||
#include <unordered_set>
|
#include <unordered_set>
|
||||||
|
|
||||||
|
#ifdef GGML_USE_KOMPUTE
|
||||||
|
#include "ggml-vulkan.h"
|
||||||
|
#endif
|
||||||
|
|
||||||
void LLModel::recalculateContext(PromptContext &promptCtx, std::function<bool(bool)> recalculate) {
|
void LLModel::recalculateContext(PromptContext &promptCtx, std::function<bool(bool)> recalculate) {
|
||||||
size_t i = 0;
|
size_t i = 0;
|
||||||
promptCtx.n_past = 0;
|
promptCtx.n_past = 0;
|
||||||
@ -174,3 +178,26 @@ std::vector<float> LLModel::embedding(const std::string &/*text*/)
|
|||||||
}
|
}
|
||||||
return std::vector<float>();
|
return std::vector<float>();
|
||||||
}
|
}
|
||||||
|
|
||||||
|
std::vector<LLModel::GPUDevice> LLModel::availableGPUDevices()
|
||||||
|
{
|
||||||
|
#if defined(GGML_USE_KOMPUTE)
|
||||||
|
std::vector<ggml_vk_device> vkDevices = ggml_vk_available_devices(0);
|
||||||
|
|
||||||
|
std::vector<LLModel::GPUDevice> devices;
|
||||||
|
for(const auto& vkDevice : vkDevices) {
|
||||||
|
LLModel::GPUDevice device;
|
||||||
|
device.index = vkDevice.index;
|
||||||
|
device.type = vkDevice.type;
|
||||||
|
device.heapSize = vkDevice.heapSize;
|
||||||
|
device.name = vkDevice.name;
|
||||||
|
device.vendor = vkDevice.vendor;
|
||||||
|
|
||||||
|
devices.push_back(device);
|
||||||
|
}
|
||||||
|
|
||||||
|
return devices;
|
||||||
|
#else
|
||||||
|
return std::vector<LLModel::GPUDevice>();
|
||||||
|
#endif
|
||||||
|
}
|
||||||
|
@ -263,12 +263,6 @@ bool ChatLLM::loadModel(const ModelInfo &modelInfo)
|
|||||||
if (m_llModelInfo.model) {
|
if (m_llModelInfo.model) {
|
||||||
// Update the settings that a model is being loaded and update the device list
|
// Update the settings that a model is being loaded and update the device list
|
||||||
MySettings::globalInstance()->setAttemptModelLoad(filePath);
|
MySettings::globalInstance()->setAttemptModelLoad(filePath);
|
||||||
std::vector<LLModel::GPUDevice> devices = m_llModelInfo.model->availableGPUDevices(0);
|
|
||||||
QVector<QString> deviceList{ "Auto" };
|
|
||||||
for (LLModel::GPUDevice &d : devices)
|
|
||||||
deviceList << QString::fromStdString(d.name);
|
|
||||||
deviceList << "CPU";
|
|
||||||
MySettings::globalInstance()->setDeviceList(deviceList);
|
|
||||||
|
|
||||||
// Pick the best match for the device
|
// Pick the best match for the device
|
||||||
QString actualDevice = m_llModelInfo.model->implementation().buildVariant() == "metal" ? "Metal" : "CPU";
|
QString actualDevice = m_llModelInfo.model->implementation().buildVariant() == "metal" ? "Metal" : "CPU";
|
||||||
|
@ -1,5 +1,6 @@
|
|||||||
#include "mysettings.h"
|
#include "mysettings.h"
|
||||||
#include "modellist.h"
|
#include "modellist.h"
|
||||||
|
#include "../gpt4all-backend/llmodel.h"
|
||||||
|
|
||||||
#include <QDir>
|
#include <QDir>
|
||||||
#include <QFile>
|
#include <QFile>
|
||||||
@ -63,6 +64,13 @@ MySettings::MySettings()
|
|||||||
: QObject{nullptr}
|
: QObject{nullptr}
|
||||||
{
|
{
|
||||||
QSettings::setDefaultFormat(QSettings::IniFormat);
|
QSettings::setDefaultFormat(QSettings::IniFormat);
|
||||||
|
|
||||||
|
std::vector<LLModel::GPUDevice> devices = LLModel::availableGPUDevices();
|
||||||
|
QVector<QString> deviceList{ "Auto" };
|
||||||
|
for (LLModel::GPUDevice &d : devices)
|
||||||
|
deviceList << QString::fromStdString(d.name);
|
||||||
|
deviceList << "CPU";
|
||||||
|
setDeviceList(deviceList);
|
||||||
}
|
}
|
||||||
|
|
||||||
Q_INVOKABLE QVector<QString> MySettings::deviceList() const
|
Q_INVOKABLE QVector<QString> MySettings::deviceList() const
|
||||||
|
Loading…
Reference in New Issue
Block a user