github/gpt4all
1
0
Fork 0
mirror of https://github.com/nomic-ai/gpt4all.git synced 2025-09-06 11:00:48 +00:00
Code Issues Packages Projects Releases Wiki Activity

Dlopen better implementation management (Version 2)

Browse Source
This commit is contained in:
niansa
2023-05-31 21:37:25 +02:00
committed by Adam Treat
parent 92407438c8
commit a3d08cdcd5
7 changed files with 80 additions and 47 deletions
Download Patch File Download Diff File Expand all files Collapse all files

52
gpt4all-backend/llmodel.h
View File

@@ -1,23 +1,35 @@
#ifndef LLMODEL_H
#define LLMODEL_H
#include "dlhandle.h"
#include <string>
#include <functional>
#include <vector>
#include <string_view>
#include <fstream>
#include <cstdint>
class LLModel {
public:
explicit LLModel() {}
virtual ~LLModel() {}
class Implementation {
LLModel *(*construct_)();
static LLModel *construct(const std::string &modelPath, std::string buildVariant = "default");
public:
Implementation(Dlhandle&&);
virtual bool loadModel(const std::string &modelPath) = 0;
virtual bool isModelLoaded() const = 0;
virtual size_t stateSize() const { return 0; }
virtual size_t saveState(uint8_t */*dest*/) const { return 0; }
virtual size_t restoreState(const uint8_t */*src*/) { return 0; }
static bool isImplementation(const Dlhandle&);
std::string_view modelType, buildVariant;
bool (*magicMatch)(std::ifstream& f);
Dlhandle dlhandle;
LLModel *construct() const {
auto fres = construct_();
fres->implementation = this;
return fres;
}
};
struct PromptContext {
std::vector<float> logits; // logits of current context
std::vector<int32_t> tokens; // current tokens in the context window
@@ -31,8 +43,17 @@ public:
float repeat_penalty = 1.10f;
int32_t repeat_last_n = 64; // last n tokens to penalize
float contextErase = 0.75f; // percent of context to erase if we exceed the context
// window
// window
};
explicit LLModel() {}
virtual ~LLModel() {}
virtual bool loadModel(const std::string &modelPath) = 0;
virtual bool isModelLoaded() const = 0;
virtual size_t stateSize() const { return 0; }
virtual size_t saveState(uint8_t */*dest*/) const { return 0; }
virtual size_t restoreState(const uint8_t */*src*/) { return 0; }
virtual void prompt(const std::string &prompt,
std::function<bool(int32_t)> promptCallback,
std::function<bool(int32_t, const std::string&)> responseCallback,
@@ -41,15 +62,18 @@ public:
virtual void setThreadCount(int32_t /*n_threads*/) {}
virtual int32_t threadCount() const { return 1; }
const char *getModelType() const {
return modelType;
const Implementation& getImplementation() const {
return *implementation;
}
static const std::vector<Implementation>& getImplementationList();
static const Implementation *getImplementation(std::ifstream& f, const std::string& buildVariant);
static LLModel *construct(const std::string &modelPath, std::string buildVariant = "default");
protected:
const Implementation *implementation;
virtual void recalculateContext(PromptContext &promptCtx,
std::function<bool(bool)> recalculate) = 0;
const char *modelType;
};
#endif // LLMODEL_H