File tree 2 files changed +6
-28
lines changed
2 files changed +6
-28
lines changed Original file line number Diff line number Diff line change @@ -4,9 +4,9 @@ services:
4
4
image : llama-cpp-docker
5
5
environment :
6
6
- GGML_CUDA_NO_PINNED=1
7
- - LLAMA_CTX_SIZE =2048
8
- - LLAMA_MODEL =/models/Meta-Llama-3.1-8B-Instruct-Q5_K_M.gguf
9
- - LLAMA_N_GPU_LAYERS =99
7
+ - LLAMA_ARG_CTX_SIZE =2048
8
+ - LLAMA_ARG_MODEL =/models/Meta-Llama-3.1-8B-Instruct-Q5_K_M.gguf
9
+ - LLAMA_ARG_N_GPU_LAYERS =99
10
10
volumes :
11
11
- ./models:/models
12
12
ports :
Original file line number Diff line number Diff line change @@ -58,35 +58,13 @@ parse_args_download_model() {
58
58
}
59
59
60
60
set_default_env_vars () {
61
- if [ -z ${LLAMA_HOST +x} ]; then
62
- export LLAMA_HOST =" 0.0.0.0"
61
+ if [ -z ${LLAMA_ARG_HOST +x} ]; then
62
+ export LLAMA_ARG_HOST =" 0.0.0.0"
63
63
fi
64
- if [ -z ${LLAMA_MODEL+x} ]; then
65
- export LLAMA_MODEL=" /models/Meta-Llama-3.1-8B-Instruct-Q5_K_M.gguf"
66
- fi
67
- }
68
-
69
- convert_llama_env_vars () {
70
- LLAMA_ARGS=$( env | grep LLAMA_ | awk ' {
71
- # for each environment variable
72
- for (n = 1; n <= NF; n++) {
73
- # replace LLAMA_ prefix with --
74
- sub("^LLAMA_", "--", $n)
75
- # find first = and split into argument name and value
76
- eq = index($n, "=")
77
- s1 = tolower(substr($n, 1, eq - 1))
78
- s2 = substr($n, eq + 1)
79
- # replace _ with - in argument name
80
- gsub("_", "-", s1)
81
- # print argument name and value
82
- print s1 " " s2
83
- }
84
- }' )
85
64
}
86
65
87
66
parse_args_download_model " $@ "
88
67
set_default_env_vars
89
- convert_llama_env_vars
90
68
91
69
set -x
92
- llama-server $LLAMA_ARGS
70
+ llama-server
You can’t perform that action at this time.
0 commit comments