char-fork/llama.cpp

git clone https://git.t4t.associates/char-fork/llama.cpp

Daniel Hanmodel: add Qwen3.8-Flash-Next (qwen4exp) (#27742)6c84c7d5d

char/aliceai-t5-moe
brancheschar/aliceai-t5-moeall refs →
1.3 KiB45 linesraw
1#pragma once
2
3#include "gguf.h"
4#include "llama.h"
5#include "llama-arch.h"
6
7#include <vector>
8
9// FIXME temporary function for better error messages
10bool llama_model_saver_supports_arch(llm_arch arch);
11
12struct llama_model_saver {
13    struct gguf_context * gguf_ctx = nullptr;
14    const bool gguf_ctx_owned;
15    const struct llama_model * model;
16    const struct LLM_KV llm_kv;
17
18    llama_model_saver(const struct llama_model * model);
19    llama_model_saver(enum llm_arch arch, struct gguf_context * gguf_ctx);
20    ~llama_model_saver();
21
22    void add_kv(enum llm_kv key, uint32_t     value);
23    void add_kv(enum llm_kv key, int32_t      value);
24    void add_kv(enum llm_kv key, uint64_t     value);
25    void add_kv(enum llm_kv key, float        value);
26    void add_kv(enum llm_kv key, bool         value);
27    void add_kv(enum llm_kv key, const char * value);
28
29    [[noreturn]]
30    void add_kv(enum llm_kv key, char value); // needed to make the template below compile
31
32    template <typename Container>
33    void add_kv(enum llm_kv key, const Container & value, bool per_layer = false);
34
35    void add_kv(enum llm_kv key, const std::vector<std::string> & value);
36
37    void add_tensor(const struct ggml_tensor * tensor);
38
39    void add_kv_from_model();
40
41    void add_tensors_from_model();
42
43    void save(const std::string & path_model);
44    void save(FILE * file);
45};