| | #!/bin/bash |
| |
|
| | N_GPU_LAYERS="${N_GPU_LAYERS:-10}" |
| | MODEL_PATH=${MODEL_PATH:-/data/models/mixtral-8x7b-v0.1.Q5_K_M.gguf} |
| | MODEL_URL="${MODEL_URL:-https://huggingface.co/TheBloke/Mixtral-8x7B-v0.1-GGUF/resolve/main/mixtral-8x7b-v0.1.Q5_K_M.gguf}" |
| | PORT="${PORT:-7860}" |
| | CONTEXT="${CONTEXT:-32768}" |
| | PARALLEL_NUMBER = "${PARALLEL_NUMBER:-4}" |
| |
|
| | |
| | print_progress() { |
| | local percentage=$1 |
| | echo "Download progress: $percentage%" |
| | } |
| |
|
| | calculate_progress() { |
| | local current=$1 |
| | local total=$2 |
| | echo "scale=2; $current / $total * 100" | bc |
| | } |
| |
|
| | |
| | if [ ! -e "$MODEL_PATH" ]; then |
| | echo "Downloading model." |
| | mkdir -p /data/models |
| | wget -x -q -O $MODEL_PATH "$MODEL_URL" |
| | echo "Model downloaded." |
| | else |
| | echo "File already exists." |
| | fi |
| | cd /app |
| | ./server -m "$MODEL_PATH" -c $CONTEXT --port $PORT --host 0.0.0.0 --n-gpu-layers $N_GPU_LAYERS --parallel 4 --cont-batching --path "/app/public" |