run.sh

#!/usr/bin/env bash
set -euo pipefail

echo "🚀 Complete AI Video Suite - Docker Container Starting..."
echo "🐳 Container: $(hostname)"
echo "👤 User: $(whoami)"
echo "🎮 GPUs: $(nvidia-smi --list-gpus | wc -l || echo '0')"

if command -v nvidia-smi >/dev/null 2>&1; then
  echo "💾 CUDA Memory:"
  nvidia-smi --query-gpu=memory.total,memory.used --format=csv,noheader,nounits | nl
fi

# Básico
export CUDA_VISIBLE_DEVICES=${CUDA_VISIBLE_DEVICES:-0,1,2,3,4,5,6,7}
export HF_HOME=${HF_HOME:-/app/model_cache}
export OUTPUT_DIR=${OUTPUT_DIR:-/app/outputs}
mkdir -p "$HF_HOME" "$OUTPUT_DIR" /app/ckpt/VINCIE-3B

# Clonar VINCIE se não existir
if [ ! -d "/app/VINCIE" ]; then
  echo "📦 Clonando VINCIE..."
  git clone https://github.com/ByteDance-Seed/VINCIE /app/VINCIE
fi

# Baixar modelo
python3 - <<'PY'
from huggingface_hub import snapshot_download
import os
save_dir = '/app/ckpt/VINCIE-3B'
os.makedirs(save_dir, exist_ok=True)
try:
    print('📥 Baixando VINCIE-3B...')
    snapshot_download(
        repo_id='ByteDance-Seed/VINCIE-3B',
        local_dir=save_dir,
        resume_download=True,
        local_dir_use_symlinks=False
    )
    print('✅ Modelo ok')
except Exception as e:
    print(f'⚠️ Download falhou: {e}')
PY


# Garantir dit.pth e vae.pth por nome
python3 - <<'PY'
import os
from huggingface_hub import hf_hub_download

save_dir = '/app/ckpt/VINCIE-3B'
os.makedirs(save_dir, exist_ok=True)

needed = ['dit.pth', 'vae.pth']
for fn in needed:
    p = os.path.join(save_dir, fn)
    if not (os.path.exists(p) and os.path.getsize(p) > 1024*1024):
        print(f'Baixando {fn}...')
        hf_hub_download(repo_id='ByteDance-Seed/VINCIE-3B', filename=fn, local_dir=save_dir)
print('✅ Checkpoints ok')
PY

# Compatibilidade com caminho relativo do repo (ckpt/VINCIE-3B)
mkdir -p /app/VINCIE/ckpt
ln -sfn /app/ckpt/VINCIE-3B /app/VINCIE/ckpt/VINCIE-3B


# Verificar flash-attn (sem reinstalar em user site)
python - <<'PY'
try:
    import torch, flash_attn
    print(f"✅ Flash-attn pronto (Torch {torch.__version__}, CUDA {torch.version.cuda})")
except Exception as e:
    print(f"⚠️ Flash-attn indisponível: {e}")
    print("➡️ Seguindo com fallback (atenção nativa do PyTorch).")
PY



# Copiar arquivos do serviço
cp /app/vincie_service.py /app/VINCIE/ 2>/dev/null || echo "vincie_service.py não encontrado"
cp /app/app.py /app/VINCIE/ 2>/dev/null || echo "app.py não encontrado"


echo "🚀 Iniciando app.py..."
python3 /app/app.py --listen --port ${PORT:-7860}