mirror of
https://github.com/ivabus/pantry
synced 2024-09-21 09:10:50 +03:00
2b06942c62
* llama.cpp, github version instead of hardcoded version * llama.cpp, check if model is specified, if yes, run it, if not, then download model * Use entrypoint for custom llama.cpp invocation * `llama.cpp` is just raw executable. This I think is our new pattern. * To run chat use the entrypoint: `pkgx +brewkit -- run llama.cpp` Co-authored-by: James Reynolds <magnsuviri@me.com> Co-authored-by: Max Howell <mxcl@me.com>
33 lines
866 B
Bash
Executable file
33 lines
866 B
Bash
Executable file
#!/usr/bin/env -S pkgx +gum +aria2c bash
|
|
|
|
set -eo pipefail
|
|
test -n "$VERBOSE" && set -x
|
|
|
|
MODEL_URL="https://huggingface.co/TheBloke/dolphin-2.1-mistral-7B-GGUF/resolve/main/dolphin-2.1-mistral-7b.Q4_0.gguf"
|
|
MODEL_FILENAME=$(basename "$MODEL_URL")
|
|
MODEL_DIR="${XDG_DATA_HOME:-$HOME/.local/share}"/llama.cpp
|
|
|
|
if [ ! -f "$MODEL_DIR/$MODEL_FILENAME" ]; then
|
|
gum format <<EoMD
|
|
# downloading $MODEL_FILENAME
|
|
models will be placed: \`$PWD\`
|
|
> this may take a a few minutes…
|
|
EoMD
|
|
echo #spacer
|
|
mkdir -p "$MODEL_DIR"
|
|
aria2c "$MODEL_URL" --dir="$MODEL_DIR"
|
|
gum format "# All done!"
|
|
echo #spacer
|
|
fi
|
|
|
|
D="$(cd "$(dirname "$0")" && pwd)"
|
|
|
|
exec "$D"/bin/llama.cpp \
|
|
--model "$MODEL_DIR/$MODEL_FILENAME" \
|
|
-n 256 \
|
|
--repeat_penalty 1.0 \
|
|
--color \
|
|
--interactive \
|
|
--reverse-prompt "User:" \
|
|
--file "$D"/share/prompts/chat-with-bob.txt
|