bin/llama.sh

#!/bin/bash

function llama_init_environment {
    # if conf does not exist, create it
    if [ ! -f "$HOME/.config/llama/llama-main.conf" ]; then
        mkdir -p "$HOME/.config/llama"
        cat <<EOF > "$HOME/.config/llama/llama-main.conf"
LLAMA_TEMPERATURE=0.1
LLAMA_CONTEXT_SIZE=4096
LLAMA_REPETITION_PENALTY=1.15
LLAMA_TOP_P=0.9
LLAMA_TOP_K=20
LLAMA_TEMPLATE=chatml
LLAMA_MODEL_NAME=teknium/OpenHermes-2.5-Mistral-7B/openhermes-2.5-mistral-7b-f16.gguf
LLAMA_MODELS_PATH=$HOME/.ai/models/llama
LLAMA_PROMPT_PATH=$HOME/.local/share/llama/prompts
LLAMA_CACHE_PATH=$HOME/.cache/llama
EOF
    fi

    source $HOME/.config/llama/llama-main.conf

    if [ ! -d "$LLAMA_CACHE_PATH" ]; then
        mkdir -p "$LLAMA_CACHE_PATH"
    fi

}

function llama_interactive {
    # if 4 arguments are provided...
    if [[ $# -eq 4 ]]; then
        LLAMA_MODEL_NAME=$1
        LLAMA_TEMPLATE=$2
        LLAMA_CONTEXT_SIZE=$3
        LLAMA_TEMPERATURE=$4
    fi

    # if fewer than 4 arguments - but greater than 0 - are provided, then display error message
    if [[ $# -lt 4 ]] && [[ $# -gt 0 ]]; then
        echo "Error: 4 arguments are required."
        echo "Usage: llama <model_name> <template> <context_size> <temperature>"
        echo "Example: llama teknium/OpenHermes-2.5-Mistral-7B/openhermes-2.5-mistral-7b-f16.gguf chatml 4096 0.1"
        return
    fi

    llama \
        --n-gpu-layers 1 \
        --model "$LLAMA_MODELS_PATH/$LLAMA_MODEL_NAME" \
        --prompt-cache "$LLAMA_CACHE_PATH/${LLAMA_MODEL_NAME//[\/\.]/-}-${LLAMA_CONTEXT_SIZE}.cache" \
        --file "$(get_model_prompt $LLAMA_MODEL_NAME)" \
        --in-prefix "$(get_model_prefix $LLAMA_TEMPLATE)" \
        --in-suffix "$(get_model_suffix $LLAMA_TEMPLATE)" \
        --reverse-prompt "$(get_model_prefix $LLAMA_TEMPLATE)" \
        --reverse-prompt "<|im_end|>" \
        --threads "6" \
        --temp "$LLAMA_TEMPERATURE" \
        --top-p "$LLAMA_TOP_P" \
        --top-k "$LLAMA_TOP_K" \
        --repeat-penalty "$LLAMA_REPETITION_PENALTY" \
        --ctx-size "$LLAMA_CONTEXT_SIZE" \
        --batch-size 1024 \
        --n-predict -1 \
        --keep -1 \
        --instruct \
        --no-mmap \
        --color \
        --escape \
        --log-disable
}

function get_model_prefix {
    case $1 in
        chatml|*dolphin*)
            printf "<|im_start|>user\\\n"
            ;;
        *guanaco*)
            printf "### Human: "
            ;;
        *alpaca*|*upstage*|*airoboros*|*hermes*)
            printf "### Instruction: "
            ;;
        *vicuna*|*wizardlm*|*samantha*|*scarlett*)
            printf "USER: "
            ;;
        *based*|*yi*)
            printf "Human: "
            ;;
        *orca*)
            printf "### User: "
            ;;
        *)
            printf "Input: "
            ;;
    esac
}

function get_model_suffix {
    case $1 in
        chatml|*dolphin*)
            printf "<|im_end|>\n<|im_start|>assistant\\\n"
            ;;
        *guanaco*)
            printf "### Assistant:"
            ;;
        *alpaca*|*upstage*|*airoboros*|*hermes*)
            printf "### Response:"
            ;;
        *vicuna*|*wizardlm*|*samantha*|*scarlett*)
            printf "ASSISTANT:"
            ;;
        *based*|*yi*)
            printf "Assistant:"
            ;;
        *orca*)
            printf "### Response:"
            ;;
        *)
            printf "Output:"
            ;;
    esac
}

function get_model_prompt {
    case $1 in
        *guanaco*)
            echo "$LLAMA_PROMPT_PATH/guanaco.txt"
            ;;
        *alpaca*|*upstage*|*airoboros*|*hermes*)
            echo "$LLAMA_PROMPT_PATH/alpaca.txt"
            ;;
        *dolphin*)
            echo "$LLAMA_PROMPT_PATH/chatml-gpt.txt"
            ;;
        *vicuna*|*wizardlm*)
            echo "$LLAMA_PROMPT_PATH/vicuna-v11.txt"
            ;;
        *scarlett*)
            echo "$LLAMA_PROMPT_PATH/scarlett.txt"
            ;;
        *samantha*)
            echo "$LLAMA_PROMPT_PATH/samantha.txt"
            ;;
        *based*)
            echo "$LLAMA_PROMPT_PATH/based.txt"
            ;;
        *yi*)
            echo "$LLAMA_PROMPT_PATH/yi.txt"
            ;;
        *orca*)
            echo "$LLAMA_PROMPT_PATH/orca.txt"
            ;;
        *)
            echo "$LLAMA_PROMPT_PATH/idm-gpt-lite.txt"
            ;;
    esac
}

llama_init_environment
llama_interactive $*