start.sh

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

: "${APP_DIR:=/app}"

export PYTORCH_CUDA_ALLOC_CONF="expandable_segments:True,max_split_size_mb:128"
export CUDA_MODULE_LOADING="LAZY"

echo "======================================================="
echo " VINCIE - Start (VINCIE-3B, 8x L40S)"
echo "======================================================="

# 1) Builder (garante Apex/Flash e deps CUDA)
echo "🛠️  Iniciando o builder.sh para compilar/instalar dependências CUDA..."
if [ -f "/app/builder.sh" ]; then
  /bin/bash /app/builder.sh
  echo "✅ Builder finalizado."
else
  echo "⚠️  Aviso: builder.sh não encontrado. Pulando etapa de compilação de dependências."
fi

# 2) Env da UI
export GRADIO_SERVER_NAME="0.0.0.0"
export GRADIO_SERVER_PORT="${PORT:-7860}"
export GRADIO_ENABLE_QUEUE="True"

# 3) Preflight: repo + snapshot + symlinks idempotentes
python3 - <<'PY'
import os, sys, subprocess
from pathlib import Path
from services.vincie import VincieService

repo_dir = Path("/app/VINCIE")
ckpt_repo = repo_dir / "ckpt" / "VINCIE-3B"
ckpt_app  = Path("/app/ckpt") / "VINCIE-3B"

def ensure_symlink(link: Path, target: Path):
    if link.is_symlink():
        # relinka se o alvo mudou
        if link.resolve() != target:
            link.unlink()
            link.symlink_to(target, target_is_directory=True)
    elif link.exists():
        # se for pasta/arquivo, remove para padronizar como symlink
        subprocess.run(["rm","-rf",str(link)], check=True)
        link.symlink_to(target, target_is_directory=True)
    else:
        link.parent.mkdir(parents=True, exist_ok=True)
        link.symlink_to(target, target_is_directory=True)

# 3.1 baixa repo + snapshot
svc = VincieService()
svc.ensure_repo()
svc.ensure_model()
snapshot = Path(str(svc.ckpt_dir))

# 3.2 cria symlink dentro do repo e também em /app/ckpt
ensure_symlink(ckpt_repo, snapshot)
ensure_symlink(ckpt_app,  snapshot)

# 3.3 valida itens essenciais
need = [snapshot/"dit.pth", snapshot/"vae.pth", snapshot/"llm14b"]
missing = [str(p) for p in need if not p.exists()]
if missing:
    raise SystemExit(f"[preflight] faltam itens no snapshot: {missing}")

# 3.4 symlink opcional de models (heranças YAML)
models_link = Path("/app/models")
models_src  = repo_dir / "models"
if models_src.exists() and (not models_link.exists()):
    try:
        models_link.symlink_to(models_src, target_is_directory=True)
        print(f"[preflight] linked {models_link} -> {models_src}")
    except Exception as e:
        print("[preflight] warn: link models failed:", e)

print(f"[preflight] OK: repo={repo_dir}, ckpt(link)={ckpt_repo} -> {snapshot}")
PY

# 4) PYTHONPATH (imports upstream)
export VINCIE_DIR="${VINCIE_DIR:-/app/VINCIE}"
: "${PYTHONPATH:=}"
if [ -n "${PYTHONPATH}" ]; then
  export PYTHONPATH="${VINCIE_DIR}:${PYTHONPATH}"
else
  export PYTHONPATH="${VINCIE_DIR}"
fi

# 5) Diagnóstico mínimo
echo "🔎 ckpt (repo):"
ls -la /app/VINCIE/ckpt || true
ls -la /app/VINCIE/ckpt/VINCIE-3B | head -n 20 || true
echo "🔎 ckpt (/app):"
ls -la /app/ckpt || true
ls -la /app/ckpt/VINCIE-3B | head -n 20 || true

# 6) Sobe a UI
echo "🚀 Iniciando a interface web VINCIE (app_vince.py)..."
cd "$APP_DIR"
python3 /app/app_vince.py