File size: 9,533 Bytes
5b4f66e | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 | import os
import sys
import subprocess
import shutil
import venv
import json
from pathlib import Path
ROOT = Path(__file__).resolve().parent
VENV_DIR = ROOT / ".rar_env"
REPO_DIR = ROOT / "1d-tokenizer"
OUT_DIR = ROOT / "outputs_rar"
WEIGHT_DIR = ROOT / "weights"
# Defaults for direct run (no terminal args needed)
DEFAULT_CLASS_ID = 207
DEFAULT_CLASS_IDS = [1,3,5,9] # e.g., [207, 282, 404] to generate multiple by default
DEFAULT_NUM_IMAGES = 1
DEFAULT_RAR_SIZE = "rar_xl" # one of: rar_b, rar_l, rar_xl, rar_xxl
def run(cmd, cwd=None, env=None, check=True, quiet=False):
# Nicely print the command without dumping large inline code blobs
display = cmd[:]
if "-c" in display:
try:
i = display.index("-c")
if i + 1 < len(display):
display[i + 1] = "<inline>"
except ValueError:
pass
print(f"[run] {' '.join(display)}")
stdout = subprocess.DEVNULL if quiet else None
stderr = subprocess.DEVNULL if quiet else None
return subprocess.run(cmd, cwd=cwd, env=env, check=check, stdout=stdout, stderr=stderr)
def ensure_venv() -> Path:
"""Create a local venv if missing and return its python path."""
if not VENV_DIR.exists():
print(f"[setup] Creating venv at {VENV_DIR}")
builder = venv.EnvBuilder(with_pip=True)
builder.create(VENV_DIR)
# Determine python executable inside venv (Windows/Linux)
if os.name == 'nt':
py = VENV_DIR / "Scripts" / "python.exe"
else:
py = VENV_DIR / "bin" / "python"
return py
def in_venv() -> bool:
return sys.prefix != getattr(sys, "base_prefix", sys.prefix)
def install_requirements(venv_python: Path):
# Upgrade pip tooling first
run([str(venv_python), "-m", "pip", "install", "--upgrade", "pip", "setuptools", "wheel", "-q"], quiet=True)
# Clone repo first so we can install its requirements
if not REPO_DIR.exists():
print(f"[setup] Cloning bytedance/1d-tokenizer into {REPO_DIR}")
run(["git", "clone", "https://github.com/bytedance/1d-tokenizer", str(REPO_DIR)])
else:
print(f"[setup] Repo exists, pulling latest...")
run(["git", "pull", "--ff-only"], cwd=str(REPO_DIR))
# Install repo requirements
req = REPO_DIR / "requirements.txt"
deps_marker = VENV_DIR / ".deps_installed"
if req.exists():
if not deps_marker.exists():
print("[setup] Installing repo requirements (first time)")
run([str(venv_python), "-m", "pip", "install", "-r", str(req), "-q"], quiet=True)
# Ensure diffusers only if needed
cp = run([str(venv_python), "-c", "import diffusers"], check=False, quiet=True)
if cp.returncode != 0:
run([str(venv_python), "-m", "pip", "install", "diffusers<0.32", "-q"], quiet=True)
deps_marker.write_text("ok")
else:
print("[setup] Requirements already installed; skipping")
else:
print("[warn] requirements.txt not found; installing minimal deps")
run([str(venv_python), "-m", "pip", "install",
"torch>=2.0.0", "torchvision", "omegaconf", "transformers", "timm",
"open_clip_torch", "einops", "scipy", "pillow", "accelerate",
"gdown", "huggingface-hub", "wandb", "torch-fidelity", "torchinfo", "webdataset", "-q"], quiet=True)
def reexec_in_venv(venv_python: Path):
# Re-exec this script inside the venv
env = os.environ.copy()
env["RAR_BOOTSTRAPPED"] = "1"
cmd = [str(venv_python), str(Path(__file__).resolve())] + sys.argv[1:]
run(cmd, env=env)
sys.exit(0)
def hf_download(venv_python: Path, repo_id: str, filename: str, local_dir: Path) -> Path:
local_dir.mkdir(parents=True, exist_ok=True)
code = f"""
import sys
from pathlib import Path
from huggingface_hub import hf_hub_download
path = hf_hub_download(repo_id={repo_id!r}, filename={filename!r}, local_dir={str(local_dir)!r})
print(path)
"""
cp = subprocess.run([str(venv_python), "-c", code], stdout=subprocess.PIPE, text=True, check=True)
p = Path(cp.stdout.strip())
if not p.exists():
raise RuntimeError(f"Download failed for {repo_id}/{filename}")
return p
def generate_imagenet_class(venv_python: Path, class_id: int, rar_size: str = "rar_xl", num_images: int = 1):
OUT_DIR.mkdir(parents=True, exist_ok=True)
WEIGHT_DIR.mkdir(parents=True, exist_ok=True)
# Ensure weights are present
print("[weights] Downloading tokenizer and RAR weights if missing...")
tok_path = hf_download(venv_python, "fun-research/TiTok", "maskgit-vqgan-imagenet-f16-256.bin", WEIGHT_DIR)
rar_bin = f"{rar_size}.bin"
rar_path = hf_download(venv_python, "yucornetto/RAR", rar_bin, WEIGHT_DIR)
# Execute generation inline inside the venv
code = f"""
import sys
from pathlib import Path
import traceback
REPO_DIR = Path({str(REPO_DIR)!r})
WEIGHT_DIR = Path({str(WEIGHT_DIR)!r})
OUT_DIR = Path({str(OUT_DIR)!r})
try:
import torch
from PIL import Image
if str(REPO_DIR) not in sys.path:
sys.path.insert(0, str(REPO_DIR))
import demo_util
from modeling.titok import PretrainedTokenizer
from modeling.rar import RAR
cfg_map = {{
'rar_xl': dict(hidden_size=1280, layers=32, heads=16, mlp=5120),
}}
rar_size = {rar_size!r}
assert rar_size in cfg_map, f"Unsupported rar size: {{rar_size}}"
config = demo_util.get_config(str(REPO_DIR / 'configs' / 'training' / 'generator' / 'rar.yaml'))
config.experiment.generator_checkpoint = str(WEIGHT_DIR / f"{{rar_size}}.bin")
config.model.generator.hidden_size = cfg_map[rar_size]['hidden_size']
config.model.generator.num_hidden_layers = cfg_map[rar_size]['layers']
config.model.generator.num_attention_heads = cfg_map[rar_size]['heads']
config.model.generator.intermediate_size = cfg_map[rar_size]['mlp']
config.model.vq_model.pretrained_tokenizer_weight = str(WEIGHT_DIR / 'maskgit-vqgan-imagenet-f16-256.bin')
device = 'cuda' if torch.cuda.is_available() else 'cpu'
tokenizer = PretrainedTokenizer(config.model.vq_model.pretrained_tokenizer_weight)
generator = RAR(config)
generator.load_state_dict(torch.load(config.experiment.generator_checkpoint, map_location='cpu'))
generator.eval(); generator.requires_grad_(False); generator.set_random_ratio(0)
tokenizer.to(device)
generator.to(device)
cls_id = int({class_id})
num_images = int({num_images})
OUT_DIR.mkdir(parents=True, exist_ok=True)
for i in range(num_images):
imgs = demo_util.sample_fn(
generator=generator,
tokenizer=tokenizer,
labels=[cls_id],
randomize_temperature=1.02,
guidance_scale=6.9,
guidance_scale_pow=1.5,
device=device,
)
Image.fromarray(imgs[0]).save(OUT_DIR / f'rar_{{rar_size}}_cls{{cls_id}}_{{i}}.png')
print('DONE')
except Exception:
print('[ERROR] Generation failed:')
traceback.print_exc()
raise
"""
run([str(venv_python), "-c", code])
def parse_args():
import argparse
p = argparse.ArgumentParser(description="RAR-XL one-shot setup and sampling")
p.add_argument("--class_id", type=int, default=DEFAULT_CLASS_ID, help="ImageNet-1K class id [0..999]")
p.add_argument("--rar_size", type=str, default=DEFAULT_RAR_SIZE, help="RAR model variant (fixed to rar_xl)")
p.add_argument("--num_images", type=int, default=DEFAULT_NUM_IMAGES, help="Number of images to generate")
p.add_argument("--class_ids", type=int, nargs='+', help="Generate for multiple class ids [0..999]")
args = p.parse_args()
# Enforce XL regardless of user input
args.rar_size = "rar_xl"
# Optional: supply class IDs via env var or classes.txt without terminal args
if args.class_ids is None:
env_cls = os.environ.get("RAR_CLASS_IDS")
if env_cls:
try:
args.class_ids = [int(x.strip()) for x in env_cls.split(',') if x.strip()]
except Exception:
args.class_ids = None
if args.class_ids is None:
classes_file = ROOT / "classes.txt"
if classes_file.exists():
try:
raw = classes_file.read_text()
args.class_ids = [int(x) for x in raw.replace('\n', ' ').split() if x.strip()]
except Exception:
args.class_ids = None
if args.class_ids is None and DEFAULT_CLASS_IDS:
args.class_ids = list(DEFAULT_CLASS_IDS)
return args
def main():
args = parse_args()
# Phase 1: ensure venv and requirements
if not in_venv() and os.environ.get("RAR_BOOTSTRAPPED") != "1":
vpy = ensure_venv()
install_requirements(vpy)
reexec_in_venv(vpy)
return
# Phase 2: already in venv — clone if needed (done in install), then generate
# Ensure repo exists (in case venv already existed but repo missing)
if not REPO_DIR.exists():
run(["git", "clone", "https://github.com/bytedance/1d-tokenizer", str(REPO_DIR)])
vpy = Path(sys.executable)
if args.class_ids:
for cid in args.class_ids:
generate_imagenet_class(vpy, class_id=int(cid), rar_size=args.rar_size, num_images=args.num_images)
else:
generate_imagenet_class(vpy, class_id=args.class_id, rar_size=args.rar_size, num_images=args.num_images)
print(f"[done] Images saved to {OUT_DIR}")
if __name__ == "__main__":
main()
|