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
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
|
#include <stdio.h>
#include <stdbool.h>
#include <ctype.h>
#include <err.h>
#include <errno.h>
#include <signal.h>
#include <string.h>
#include <unistd.h>
#include <sys/un.h>
#include <sys/stat.h>
#include <sys/socket.h>
#include "llm.h"
#define SOCK_PATH "/var/run/lex.sock"
#define MODEL_PATH "qwen2.5-3b-instruct-q4_k_m.gguf"
static void usage(const char *bin)
{
errx(1, "usage: %s [-m model_path] [-d] [prompt]", bin);
}
static void loop(struct llama_model *model, int lsock)
{
int csock, c;
FILE *cfp;
size_t len;
char line[4096];
for (;;) {
csock = accept(lsock, NULL, NULL);
if (csock == -1) {
if (errno != EINTR)
warn("accept()");
continue;
}
if (!(cfp = fdopen(csock, "r+"))) {
warn("cfp fdopen()");
close(csock);
continue;
}
if (fgets(line, sizeof(line), cfp) != NULL) {
len = strcspn(line, "\n");
if (line[len] == '\0' && len == sizeof(line) - 1) {
warnx("request too large");
// drain and skip
while ((c = fgetc(cfp)) != '\n' && c != EOF)
;
fclose(cfp);
continue;
}
line[len] = '\0';
if (line[0] != '\0')
llm_process_request(model, line, cfp);
}
fclose(cfp);
}
}
int sock_init(void)
{
int sock;
struct sockaddr_un addr;
sock = socket(AF_UNIX, SOCK_STREAM, 0);
if (sock == -1)
err(1, "socket()");
memset(&addr, 0, sizeof(addr));
addr.sun_family = AF_UNIX;
strlcpy(addr.sun_path, SOCK_PATH, sizeof(addr.sun_path));
unlink(SOCK_PATH); /* remove stale socket from a previous run */
if (bind(sock, (struct sockaddr *)&addr, sizeof(addr)) == -1)
err(1, "bind()");
if (listen(sock, 1) == -1) /* backlog = 1 */
err(1, "listen()");
return sock;
}
int main(int argc , char *argv[])
{
int daemon_mode, sock, opt;
const char *prog;
const char *prompt;
const char *model_path;
struct llama_model *model;
if (pledge("stdio rpath wpath cpath unix", NULL) == -1)
err(1, "pledge failed");
sock = -1;
daemon_mode = 0;
prompt = NULL;
prog = argv[0];
model_path = MODEL_PATH;
while ((opt = getopt(argc, argv, "dm:")) != -1) {
switch (opt) {
case 'd':
daemon_mode = 1;
break;
case 'm':
model_path = optarg;
break;
default:
usage(prog);
}
}
argc -= optind;
argv += optind;
if (daemon_mode) {
if (argc != 0) // -d takes no positional args
usage(prog);
} else {
if (argc != 1) // normal mode requires one prompt
usage(prog);
prompt = argv[0];
}
if (unveil(model_path, "r") == -1)
err(1, "unveil %s failed", model_path);
if (daemon_mode && unveil(SOCK_PATH, "rwc") == -1)
err(1, "unveil %s failed", SOCK_PATH);
if (unveil(NULL, NULL) == -1)
err(1, "unveil lock failed");
if (daemon_mode) {
sock = sock_init();
if (pledge("stdio rpath unix", NULL) == -1)
err(1, "Failed to narrow pledge after sock_init()");
}
model = llm_init(model_path);
if (!model)
errx(1, "Failed to load model");
if (daemon_mode)
loop(model, sock); /* doesn't return */
if (pledge("stdio", NULL) == -1)
err(1, "Failed to narrow pledge for normal mode");
llm_process_request(model, prompt, stdout);
llm_free(model);
return 0;
}
|
