kobold.sh

#!/bin/bash
#source ~/koboldcpp/venv/bin/activate
model_dir="/mnt/Orlando/gguf"
model_files=("$model_dir"/*.gguf)
if [ ${#model_files[@]} -eq 0 ]; then
    echo "No .gguf files found in the specified directory."
    exit 1
fi
echo "Select a model file:"
select model_file in "${model_files[@]}"; do
    if [ -n "$model_file" ]; then
        break
    else
        echo "Invalid selection. Please try again."
    fi
done
echo "Select context size (Press Enter to default to 8192):"

options=("256" "512" "1024" "2048" "3072" "4096" "6144" "8192" "12288" "16384" "24576" "32768" "49152" "65536")

PS3="Context: "

select opt in "${options[@]}" ""; do
    case $REPLY in
        "")
            context="8192"
            echo "No context size selected. Defaulting to 8192."
            break
            ;;
        *[0-9]*)
            if [[ $REPLY -ge 1 && $REPLY -le ${#options[@]} ]]; then
                context=${options[$REPLY-1]}
                echo "Selected context size: $context"
                break
            else
                echo "Invalid option. Please try again."
            fi
            ;;
        *)
            echo "Invalid input. Please enter a number."
            ;;
    esac
done

read -p "Do you want to change other parameters? (y/[n]): " change_params
if [[ "$change_params" == "y" ]]; then

while true; do
    read -p "Enter tensor split [disabled]: " tensor_response
    if [[ "$tensor_response" == "" ]]; then
        tensor_split=""
        break
    elif [[ "$tensor_response" =~ ^[0-9]+(\.[0-9]+)?(\ [0-9]+(\.[0-9]+)?)*$ ]]; then
        tensor_split=("--tensor_split $tensor_response")
        break
    else
        echo "Invalid response. Please enter a valid tensor split value (integers or floats separated by spaces)."
    fi
done

read -p "Enter gpu layers [999]: " layers_response
layers=${layers_response:-999}

read -p "Disable rowsplit? (y/[n]): " rowsplit_response
rowsplit=$([ "$rowsplit_response" == "y" ] && echo "" || echo "")

read -p "Preload (y/[n]): " preload_response
preload=$([ "$preload_response" == "y" ] && echo "--preloadstory /mnt/Orlando/saved_settings.json" || echo "")

#read -p "Enable txt-to-image? (y/[n]): " sd_response
#sd=$([ "$sd_response" == "y" ] && echo "--sdconfig /mnt/Orlando/Models/Stable-diffusion/v1-5-pruned-emaonly.safetensors" || echo "")

#read -p "Enable debug mode? (y/[n]): " debug_response
#debug=$([ "$debug_response" == "y" ] && echo "--debug" || echo "")

read -p "Enable non-quiet mode? (y/[n]): " quiet_response
quiet=$([ "$quiet_response" == "y" ] && echo "" || echo "--quiet")

#read -p "Enable tunnel? (y/[n]): " tunnel_response
#tunnel=$([ "$tunnel_response" == "y" ] && echo "--remotetunnel" || echo "")
else


layers=999
preload=""
sd=""
rowsplit="rowsplit"
debug=""
quiet="--quiet"
tunnel=""
fi

debug=""
sd=""
tunnel=""

additional_flags=("$@")

python ~/kbnew/koboldcpp/koboldcpp.py --usecublas $rowsplit --port 5001 --contextsize "$context" --gpulayers $layers --multiuser --model "$model_file" $quiet $tensor_split $preload $tunnel $sd $debug "${additional_flags[@]}"