#!/usr/bin/env bash
# Ling 3.1 Flash lokal ausprobieren – Linux
# Automatisch erzeugt von Tokenstrom (tokenstrom.com).
# Aufruf: bash try_linux.sh            # Ollama (Standard)
#        MODE=hf bash try_linux.sh    # Python/transformers
set -e
cd "$(dirname "$0")"
echo "== Ling 3.1 Flash =="
if command -v nvidia-smi >/dev/null 2>&1; then nvidia-smi --query-gpu=name,memory.total --format=csv,noheader;
else echo "No NVIDIA GPU found – running on CPU (slow)"; fi
echo "WARNUNG: Groß (572,1 B parameters, MoE): running entirely on the GPU takes a lot of VRAM (video memory). With MoE offloading, a regular graphics card plus plenty of RAM is enough (see table). Colab Free isn't enough."

if [ "${MODE:-ollama}" = "ollama" ]; then
  echo "== Ollama =="
  if ! command -v ollama >/dev/null 2>&1; then
    command -v zstd >/dev/null 2>&1 || sudo apt-get install -y zstd 2>/dev/null || sudo dnf install -y zstd 2>/dev/null || true
    curl -fsSL https://ollama.com/install.sh | sh
  fi
  curl -s http://localhost:11434/api/tags >/dev/null 2>&1 || (ollama serve >/dev/null 2>&1 &)
  sleep 3
  MODEL="${MODEL:-}"
  MODEL="${MODEL:-hf.co/Baekpica/Ling-3.1-flash-VQ-GGUF}"
  echo "Downloading $MODEL (the first time this takes a few minutes, depending on size) ..."
  ollama pull "$MODEL"
  echo ""
  echo "Done! Now type your question and press Enter. Quit with /bye"
  exec ollama run "$MODEL"
fi

echo "== Python / Hugging Face =="
python3 -m venv .venv-ling-3-1-flash
source .venv-ling-3-1-flash/bin/activate
pip install -q -U pip
if command -v nvidia-smi >/dev/null 2>&1; then
  pip install -q -U torch --index-url https://download.pytorch.org/whl/cu124
  pip install -q -U transformers accelerate bitsandbytes
else
  pip install -q -U torch --index-url https://download.pytorch.org/whl/cpu
  pip install -q -U transformers accelerate
fi

python3 demo.py