#!/usr/bin/env bash
# Ling-3.0-flash-Sante lokal ausprobieren – macOS
# Automatisch erzeugt von Tokenstrom (tokenstrom.com).
# Aufruf: bash try_mac.sh            # Ollama (Standard)
#        MODE=hf bash try_mac.sh    # Python/transformers
set -e
cd "$(dirname "$0")"
echo "== Ling-3.0-flash-Sante =="
echo "WARNUNG: Groß (124 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 [ "${MODE:-ollama}" = "ollama" ]; then
  echo "== Ollama =="
  if ! command -v ollama >/dev/null 2>&1; then
    if command -v brew >/dev/null 2>&1; then brew install ollama; else
      echo "Please install the Ollama app first: https://ollama.com/download/mac"; exit 1; fi
  fi
  curl -s http://localhost:11434/api/tags >/dev/null 2>&1 || (ollama serve >/dev/null 2>&1 &)
  sleep 3
  MODEL="${MODEL:-}"
  # Pick the right variant for your hardware (to choose your own:  MODEL=<name> bash try_mac.sh)
  if [ -z "${MODEL:-}" ]; then
  MEM_GB=$(( $(sysctl -n hw.memsize) / 1073741824 * 7 / 10 ))   # the GPU can use about 70% of memory
  MODEL=$(printf '%s\n' "123.0 hf.co/AtomicChat/Ling-3.0-flash-Sante-GGUF:Q6_K 0
151.5 hf.co/AtomicChat/Ling-3.0-flash-Sante-GGUF:Q8_0 1" | awk -v mem="$MEM_GB" '
    { need[NR]=$1+0; ref[NR]=$2; def[NR]=$3+0 }
    END { mem=mem+0; for (i=1;i<=NR;i++) if (def[i]==1 && need[i]<=mem) { print ref[i]; exit }
          best=ref[1]; for (i=1;i<=NR;i++) if (need[i]<=mem) best=ref[i]; print best }')
  echo "Available memory: approx. ${MEM_GB} GB  ->  selected variant: $MODEL"
  fi

  MODEL="${MODEL:-hf.co/AtomicChat/Ling-3.0-flash-Sante-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-3-0-flash-sante
source .venv-ling-3-0-flash-sante/bin/activate
pip install -q -U pip
pip install -q -U torch transformers accelerate

python3 demo.py