Spaces:
Paused
Paused
Update start.sh
Browse files
start.sh
CHANGED
|
@@ -3,81 +3,96 @@ set -euo pipefail
|
|
| 3 |
|
| 4 |
: "${APP_DIR:=/app}"
|
| 5 |
|
| 6 |
-
export PYTORCH_CUDA_ALLOC_CONF="expandable_segments:True,max_split_size_mb:128"
|
| 7 |
-
export CUDA_MODULE_LOADING="LAZY"
|
| 8 |
-
|
| 9 |
echo "======================================================="
|
| 10 |
echo " VINCIE - Start (VINCIE-3B, 8x L40S)"
|
| 11 |
echo "======================================================="
|
| 12 |
|
| 13 |
-
# 1) Builder (
|
| 14 |
-
echo "🛠️ Iniciando o builder.sh
|
| 15 |
if [ -f "/app/builder.sh" ]; then
|
| 16 |
/bin/bash /app/builder.sh
|
| 17 |
echo "✅ Builder finalizado."
|
| 18 |
else
|
| 19 |
-
echo "⚠️ Aviso: builder.sh não encontrado. Pulando
|
| 20 |
-
fi
|
| 21 |
|
| 22 |
# 2) Env da UI
|
| 23 |
export GRADIO_SERVER_NAME="0.0.0.0"
|
| 24 |
export GRADIO_SERVER_PORT="${PORT:-7860}"
|
| 25 |
-
export GRADIO_ENABLE_QUEUE="True"
|
| 26 |
|
| 27 |
-
# 3)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 28 |
python3 - <<'PY'
|
| 29 |
-
import os, sys,
|
| 30 |
from pathlib import Path
|
| 31 |
-
from
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 47 |
else:
|
| 48 |
-
|
| 49 |
-
|
| 50 |
-
|
| 51 |
-
# 3.1 baixa repo + snapshot
|
| 52 |
-
svc = VincieService()
|
| 53 |
-
svc.ensure_repo()
|
| 54 |
-
svc.ensure_model()
|
| 55 |
-
snapshot = Path(str(svc.ckpt_dir))
|
| 56 |
-
|
| 57 |
-
# 3.2 cria symlink dentro do repo e também em /app/ckpt
|
| 58 |
-
ensure_symlink(ckpt_repo, snapshot)
|
| 59 |
-
ensure_symlink(ckpt_app, snapshot)
|
| 60 |
-
|
| 61 |
-
# 3.3 valida itens essenciais
|
| 62 |
-
need = [snapshot/"dit.pth", snapshot/"vae.pth", snapshot/"llm14b"]
|
| 63 |
-
missing = [str(p) for p in need if not p.exists()]
|
| 64 |
-
if missing:
|
| 65 |
-
raise SystemExit(f"[preflight] faltam itens no snapshot: {missing}")
|
| 66 |
-
|
| 67 |
-
# 3.4 symlink opcional de models (heranças YAML)
|
| 68 |
-
models_link = Path("/app/models")
|
| 69 |
-
models_src = repo_dir / "models"
|
| 70 |
-
if models_src.exists() and (not models_link.exists()):
|
| 71 |
try:
|
| 72 |
-
|
| 73 |
-
|
| 74 |
-
|
| 75 |
-
|
| 76 |
-
|
| 77 |
-
|
| 78 |
-
|
| 79 |
-
|
| 80 |
-
#
|
|
|
|
|
|
|
| 81 |
export VINCIE_DIR="${VINCIE_DIR:-/app/VINCIE}"
|
| 82 |
: "${PYTHONPATH:=}"
|
| 83 |
if [ -n "${PYTHONPATH}" ]; then
|
|
@@ -85,16 +100,17 @@ if [ -n "${PYTHONPATH}" ]; then
|
|
| 85 |
else
|
| 86 |
export PYTHONPATH="${VINCIE_DIR}"
|
| 87 |
fi
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 88 |
|
| 89 |
-
#
|
| 90 |
-
echo "🔎
|
| 91 |
-
ls -la
|
| 92 |
-
ls -la /app/VINCIE/ckpt/VINCIE-3B | head -n 20 || true
|
| 93 |
-
echo "🔎 ckpt (/app):"
|
| 94 |
-
ls -la /app/ckpt || true
|
| 95 |
-
ls -la /app/ckpt/VINCIE-3B | head -n 20 || true
|
| 96 |
|
| 97 |
-
#
|
| 98 |
echo "🚀 Iniciando a interface web VINCIE (app_vince.py)..."
|
| 99 |
cd "$APP_DIR"
|
| 100 |
-
python3 /app/app_vince.py
|
|
|
|
| 3 |
|
| 4 |
: "${APP_DIR:=/app}"
|
| 5 |
|
|
|
|
|
|
|
|
|
|
| 6 |
echo "======================================================="
|
| 7 |
echo " VINCIE - Start (VINCIE-3B, 8x L40S)"
|
| 8 |
echo "======================================================="
|
| 9 |
|
| 10 |
+
# 1) Builder (Apex/Flash/dep. CUDA)
|
| 11 |
+
echo "🛠️ Iniciando o builder.sh..."
|
| 12 |
if [ -f "/app/builder.sh" ]; then
|
| 13 |
/bin/bash /app/builder.sh
|
| 14 |
echo "✅ Builder finalizado."
|
| 15 |
else
|
| 16 |
+
echo "⚠️ Aviso: builder.sh não encontrado. Pulando compilação."
|
| 17 |
+
fi # [file:24]
|
| 18 |
|
| 19 |
# 2) Env da UI
|
| 20 |
export GRADIO_SERVER_NAME="0.0.0.0"
|
| 21 |
export GRADIO_SERVER_PORT="${PORT:-7860}"
|
| 22 |
+
export GRADIO_ENABLE_QUEUE="True" # [file:24]
|
| 23 |
|
| 24 |
+
# 3) CKPT_ROOT externo ao repo (use volumes)
|
| 25 |
+
if [ -d "/data" ]; then
|
| 26 |
+
export CKPT_ROOT="/data/ckpt/VINCIE-3B"
|
| 27 |
+
export HF_HUB_CACHE="${HF_HUB_CACHE:-/data/.cache/huggingface/hub}"
|
| 28 |
+
else
|
| 29 |
+
export CKPT_ROOT="/app/ckpt/VINCIE-3B"
|
| 30 |
+
export HF_HUB_CACHE="${HF_HUB_CACHE:-/app/.cache/huggingface/hub}"
|
| 31 |
+
fi
|
| 32 |
+
|
| 33 |
+
# 4) Preflight: baixar snapshot COMPLETO para CKPT_ROOT
|
| 34 |
python3 - <<'PY'
|
| 35 |
+
import os, sys, shutil
|
| 36 |
from pathlib import Path
|
| 37 |
+
from huggingface_hub import snapshot_download
|
| 38 |
+
|
| 39 |
+
CKPT_ROOT = Path(os.environ["CKPT_ROOT"])
|
| 40 |
+
MODEL_REPO = os.environ.get("MODEL_REPO", "ByteDance-Seed/VINCIE-3B")
|
| 41 |
+
repo_dir = Path("/app/VINCIE")
|
| 42 |
+
gy = repo_dir / "configs" / "generate.yaml"
|
| 43 |
+
mp = repo_dir / "main.py"
|
| 44 |
+
|
| 45 |
+
def download_snapshot():
|
| 46 |
+
CKPT_ROOT.parent.mkdir(parents=True, exist_ok=True)
|
| 47 |
+
tmp = CKPT_ROOT.with_name(CKPT_ROOT.name + ".tmp")
|
| 48 |
+
if tmp.exists():
|
| 49 |
+
shutil.rmtree(tmp, ignore_errors=True)
|
| 50 |
+
snapshot_download(
|
| 51 |
+
repo_id=MODEL_REPO,
|
| 52 |
+
local_dir=str(tmp),
|
| 53 |
+
local_dir_use_symlinks=False,
|
| 54 |
+
resume_download=True,
|
| 55 |
+
)
|
| 56 |
+
if CKPT_ROOT.exists():
|
| 57 |
+
shutil.rmtree(CKPT_ROOT, ignore_errors=True)
|
| 58 |
+
tmp.rename(CKPT_ROOT)
|
| 59 |
+
|
| 60 |
+
def validate():
|
| 61 |
+
need = [CKPT_ROOT/"dit.pth", CKPT_ROOT/"vae.pth", CKPT_ROOT/"llm14b"]
|
| 62 |
+
missing = [str(p) for p in need if not p.exists()]
|
| 63 |
+
if missing:
|
| 64 |
+
raise RuntimeError(f"Snapshot incompleto: {missing}")
|
| 65 |
+
if not gy.exists() or not mp.exists():
|
| 66 |
+
raise RuntimeError("VINCIE repo inválido (faltando generate.yaml/main.py)")
|
| 67 |
+
|
| 68 |
+
# valida repo
|
| 69 |
+
if not gy.exists() or not mp.exists():
|
| 70 |
+
raise SystemExit("[preflight] Repo ausente/incompleto; clone esperado antes do start.")
|
| 71 |
+
|
| 72 |
+
try:
|
| 73 |
+
if CKPT_ROOT.exists():
|
| 74 |
+
try:
|
| 75 |
+
validate(); print(f"[preflight] CKPT_ROOT OK: {CKPT_ROOT}")
|
| 76 |
+
except Exception as e:
|
| 77 |
+
print("[preflight] validação falhou; refazendo snapshot:", e)
|
| 78 |
+
shutil.rmtree(CKPT_ROOT, ignore_errors=True)
|
| 79 |
+
download_snapshot(); validate()
|
| 80 |
else:
|
| 81 |
+
download_snapshot(); validate()
|
| 82 |
+
except Exception as e:
|
| 83 |
+
print("[preflight] falha definitiva:", e)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 84 |
try:
|
| 85 |
+
for p in (CKPT_ROOT.parent, ):
|
| 86 |
+
print(f">>> {p}")
|
| 87 |
+
if p.exists():
|
| 88 |
+
for x in sorted(p.iterdir(), key=lambda z: z.name.lower()):
|
| 89 |
+
print(" -", x.name + ("/" if x.is_dir() else ""))
|
| 90 |
+
except Exception:
|
| 91 |
+
pass
|
| 92 |
+
sys.exit(1)
|
| 93 |
+
PY # [file:24][file:25]
|
| 94 |
+
|
| 95 |
+
# 5) PYTHONPATH e includes relativos
|
| 96 |
export VINCIE_DIR="${VINCIE_DIR:-/app/VINCIE}"
|
| 97 |
: "${PYTHONPATH:=}"
|
| 98 |
if [ -n "${PYTHONPATH}" ]; then
|
|
|
|
| 100 |
else
|
| 101 |
export PYTHONPATH="${VINCIE_DIR}"
|
| 102 |
fi
|
| 103 |
+
[ -d "/app/VINCIE/models" ] && [ ! -e "/app/models" ] && ln -s /app/VINCIE/models /app/models || true
|
| 104 |
+
|
| 105 |
+
# 6) Alocador CUDA (mitigar fragmentação no cold-start)
|
| 106 |
+
export PYTORCH_CUDA_ALLOC_CONF="${PYTORCH_CUDA_ALLOC_CONF:-expandable_segments:True,max_split_size_mb:128}"
|
| 107 |
+
export CUDA_MODULE_LOADING="${CUDA_MODULE_LOADING:-LAZY}"
|
| 108 |
|
| 109 |
+
# 7) Diagnóstico leve
|
| 110 |
+
echo "🔎 Conteúdo de CKPT_ROOT:"
|
| 111 |
+
ls -la "${CKPT_ROOT}" | head -n 50 || true
|
|
|
|
|
|
|
|
|
|
|
|
|
| 112 |
|
| 113 |
+
# 8) Launch da UI
|
| 114 |
echo "🚀 Iniciando a interface web VINCIE (app_vince.py)..."
|
| 115 |
cd "$APP_DIR"
|
| 116 |
+
python3 /app/app_vince.py # [file:23]
|