llama-cpp-sys-4 0.9.0

Low Level Bindings to llama.cpp
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49

#pragma once

// Log infrastructure for the RPC backend: LOG_* are unconditional, LOG_DBG* are gated by GGML_RPC_DEBUG

#include "ggml-impl.h"

#include <cstddef>
#include <cstdlib>
#include <string>

// GGML_RPC_DEBUG verbosity: unset / 0 disabled, 1 connections and allocations, 2 per-command trace,
// 3 per-message detail. Non-numeric values fall back to 1
static inline int rpc_log_debug_level() {
    static const int level = [] {
        const char * env = std::getenv("GGML_RPC_DEBUG");
        if (env == nullptr) {
            return 0;
        }
        if (env[0] == '\0') {
            return 1;
        }
        int value = 0;
        for (const char * p = env; *p != '\0'; p++) {
            if (*p < '0' || *p > '9') {
                return 1;
            }
            value = value * 10 + (*p - '0');
            if (value > 3) {
                value = 3;
            }
        }
        return value;
    }();
    return level;
}

// Byte counts in binary units, e.g. 8208 -> "8.02 KiB". Sizes below 1 KiB stay in plain bytes
std::string rpc_log_format_size(size_t size);

// Tensor identity: "blk.0.attn_q.weight [NONE, f16, [4096, 4096, 1, 1]]"
std::string rpc_log_format_tensor(const ggml_tensor * tensor);

#define LOG_INFO(...)  GGML_LOG_INFO(__VA_ARGS__)
#define LOG_WARN(...)  GGML_LOG_WARN(__VA_ARGS__)
#define LOG_ERROR(...) GGML_LOG_ERROR(__VA_ARGS__)

#define LOG_DBG(...)  do { if (rpc_log_debug_level() >= 1) GGML_LOG_DEBUG(__VA_ARGS__); } while (0)
#define LOG_DBG2(...) do { if (rpc_log_debug_level() >= 2) GGML_LOG_DEBUG(__VA_ARGS__); } while (0)
#define LOG_DBG3(...) do { if (rpc_log_debug_level() >= 3) GGML_LOG_DEBUG(__VA_ARGS__); } while (0)