Commit
·
17a696b
1
Parent(s):
7ef77fa
update to installed paths
Browse filesllama.sh uses config defaults or cmdline
- bin/llama-finetune.sh +1 -1
- bin/llama-hf-to-q6_k.sh +1 -1
- bin/llama.sh +20 -25
bin/llama-finetune.sh
CHANGED
@@ -21,7 +21,7 @@ source $HOME/.config/llama/llama-finetune.conf
|
|
21 |
|
22 |
cd "$TRAINING_PATH" || exit
|
23 |
|
24 |
-
|
25 |
--model-base "$MODEL" \
|
26 |
--train-data "$DATA" \
|
27 |
--lora-out "lora-${DATA}.gguf" \
|
|
|
21 |
|
22 |
cd "$TRAINING_PATH" || exit
|
23 |
|
24 |
+
llama-finetune \
|
25 |
--model-base "$MODEL" \
|
26 |
--train-data "$DATA" \
|
27 |
--lora-out "lora-${DATA}.gguf" \
|
bin/llama-hf-to-q6_k.sh
CHANGED
@@ -10,7 +10,7 @@ if [ ! -f "$HOME/.config/llama/llama-hf-to-q6_k.conf" ]; then
|
|
10 |
mkdir -p "$HOME/.config/llama"
|
11 |
cat <<EOF > "$HOME/.config/llama/llama-hf-to-q6_k.conf"
|
12 |
MODELS_ROOT=$HOME/.ai/models/llama/
|
13 |
-
HF_DOWNLOADER
|
14 |
STORAGE_PATH=$HOME/scratch/hfdownloader
|
15 |
PYTHON3_EXEC=$HOME/.virtualenvs/llama.cpp/bin/python3
|
16 |
QUANTIZE_EXEC=$HOME/Work/llama.cpp/build/bin/quantize
|
|
|
10 |
mkdir -p "$HOME/.config/llama"
|
11 |
cat <<EOF > "$HOME/.config/llama/llama-hf-to-q6_k.conf"
|
12 |
MODELS_ROOT=$HOME/.ai/models/llama/
|
13 |
+
HF_DOWNLOADER=hfdownloader
|
14 |
STORAGE_PATH=$HOME/scratch/hfdownloader
|
15 |
PYTHON3_EXEC=$HOME/.virtualenvs/llama.cpp/bin/python3
|
16 |
QUANTIZE_EXEC=$HOME/Work/llama.cpp/build/bin/quantize
|
bin/llama.sh
CHANGED
@@ -1,11 +1,5 @@
|
|
1 |
#!/bin/bash
|
2 |
|
3 |
-
# if not enough arguments
|
4 |
-
if [[ $# -lt 4 ]]; then
|
5 |
-
echo "Usage: llama.sh <model> <template> <context-size> <temperature>"
|
6 |
-
exit 1
|
7 |
-
fi
|
8 |
-
|
9 |
# if conf does not exist, create it
|
10 |
if [ ! -f "$HOME/.config/llama/llama-main.conf" ]; then
|
11 |
mkdir -p "$HOME/.config/llama"
|
@@ -20,31 +14,32 @@ LLAMA_MODEL_NAME=teknium/OpenHermes-2.5-Mistral-7B/openhermes-2.5-mistral-7b-f16
|
|
20 |
EOF
|
21 |
fi
|
22 |
|
23 |
-
source $HOME/.config/llama/llama-main.conf
|
24 |
-
|
25 |
function llama_interactive {
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
|
|
|
|
|
|
32 |
|
33 |
-
|
34 |
--n-gpu-layers 1 \
|
35 |
-
--model "$(model_path $
|
36 |
-
--prompt-cache "$(cache_path $
|
37 |
-
--file "$(get_model_prompt $
|
38 |
-
--in-prefix "$(get_model_prefix $
|
39 |
-
--in-suffix "$(get_model_suffix $
|
40 |
-
--reverse-prompt "$(get_model_prefix $
|
41 |
--reverse-prompt "<|im_end|>" \
|
42 |
-
--threads "
|
43 |
-
--temp "$
|
44 |
--top-p "$LLAMA_TOP_P" \
|
45 |
--top-k "$LLAMA_TOP_K" \
|
46 |
--repeat-penalty "$LLAMA_REPETITION_PENALTY" \
|
47 |
-
--ctx-size "$
|
48 |
--batch-size 1024 \
|
49 |
--n-predict -1 \
|
50 |
--keep -1 \
|
@@ -158,4 +153,4 @@ function get_model_prompt {
|
|
158 |
fi
|
159 |
}
|
160 |
|
161 |
-
llama_interactive
|
|
|
1 |
#!/bin/bash
|
2 |
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
# if conf does not exist, create it
|
4 |
if [ ! -f "$HOME/.config/llama/llama-main.conf" ]; then
|
5 |
mkdir -p "$HOME/.config/llama"
|
|
|
14 |
EOF
|
15 |
fi
|
16 |
|
|
|
|
|
17 |
function llama_interactive {
|
18 |
+
source $HOME/.config/llama/llama-main.conf
|
19 |
+
|
20 |
+
# if arguments are provided...
|
21 |
+
if [[ $# -eq 4 ]]; then
|
22 |
+
LLAMA_MODEL_NAME=$1
|
23 |
+
LLAMA_TEMPLATE=$2
|
24 |
+
LLAMA_CONTEXT_SIZE=$3
|
25 |
+
LLAMA_TEMPERATURE=$4
|
26 |
+
fi
|
27 |
|
28 |
+
llama \
|
29 |
--n-gpu-layers 1 \
|
30 |
+
--model "$(model_path $LLAMA_MODEL_NAME)" \
|
31 |
+
--prompt-cache "$(cache_path $LLAMA_MODEL_NAME)" \
|
32 |
+
--file "$(get_model_prompt $LLAMA_MODEL_NAME)" \
|
33 |
+
--in-prefix "$(get_model_prefix $LLAMA_TEMPLATE)" \
|
34 |
+
--in-suffix "$(get_model_suffix $LLAMA_TEMPLATE)" \
|
35 |
+
--reverse-prompt "$(get_model_prefix $LLAMA_TEMPLATE)" \
|
36 |
--reverse-prompt "<|im_end|>" \
|
37 |
+
--threads "6" \
|
38 |
+
--temp "$LLAMA_TEMPERATURE" \
|
39 |
--top-p "$LLAMA_TOP_P" \
|
40 |
--top-k "$LLAMA_TOP_K" \
|
41 |
--repeat-penalty "$LLAMA_REPETITION_PENALTY" \
|
42 |
+
--ctx-size "$LLAMA_CONTEXT_SIZE" \
|
43 |
--batch-size 1024 \
|
44 |
--n-predict -1 \
|
45 |
--keep -1 \
|
|
|
153 |
fi
|
154 |
}
|
155 |
|
156 |
+
llama_interactive $*
|