char-fork/llama.cpp

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

BartowskiMove duplicated imatrix code into single common imatrix-loader.cpp (#22445)e7bcf1c3a

char/aliceai-t5-moe
brancheschar/aliceai-t5-moeall refs →
765 B26 linesraw
1#pragma once
2
3#include <cstdint>
4#include <map>
5#include <string>
6#include <vector>
7
8inline constexpr const char * LLM_KV_IMATRIX_DATASETS    = "imatrix.datasets";
9inline constexpr const char * LLM_KV_IMATRIX_CHUNK_COUNT = "imatrix.chunk_count";
10inline constexpr const char * LLM_KV_IMATRIX_CHUNK_SIZE  = "imatrix.chunk_size";
11
12struct common_imatrix_entry {
13    std::vector<float>   sums;
14    std::vector<int64_t> counts;
15};
16
17struct common_imatrix {
18    std::map<std::string, common_imatrix_entry> entries;
19    std::vector<std::string> datasets;
20    int32_t chunk_count    = 0;
21    int32_t chunk_size     = 0;
22    bool    is_legacy      = false;
23    bool    has_metadata   = false;
24};
25
26bool common_imatrix_load(const std::string & fname, common_imatrix & imatrix);