File size: 10,751 Bytes
31dc8dc | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 | import logging
import os
import pickle
import re
from typing import Any, Optional, Tuple
import jax
import jax.numpy as jnp
from flax import serialization
from flax.training import checkpoints
from utils.logging_utils import log_for_0
def _local_path(path: str) -> str:
return os.path.abspath(os.path.expanduser(path))
def upload_output_dir_to_hf(output_dir: str, hf_repo_id: Optional[str], reason: str = "artifacts"):
if not hf_repo_id or jax.process_index() != 0:
return
folder_path = _local_path(output_dir)
if not os.path.isdir(folder_path):
log_for_0(f"HF upload skipped; output directory does not exist: {folder_path}", level=logging.WARNING)
return
try:
from huggingface_hub import HfApi
repo_id = hf_repo_id.strip("/")
api = HfApi()
api.create_repo(repo_id, repo_type="model", exist_ok=True)
log_for_0(f"Uploading {reason} to HF: {repo_id}")
api.upload_folder(repo_id=repo_id, folder_path=folder_path, repo_type="model")
log_for_0(f"Uploaded {reason} to HF: {repo_id}")
except Exception as e:
log_for_0(f"Failed to upload {reason} to HF: {e}", level=logging.WARNING)
def _split_hf_path(path: str, min_parts: int) -> Optional[Tuple[str, str]]:
if "://" in path:
return None
if path.startswith(("/", ".", "~")):
return None
if os.path.exists(_local_path(path)):
return None
parts = path.split("/")
if len(parts) < min_parts:
return None
repo_id = "/".join(parts[:2])
sub_path = "/".join(parts[2:])
return repo_id, sub_path
def save_checkpoint(state: Any, output_dir: str, step: int, hf_repo_id: str = None):
"""Save model checkpoint locally, optionally mirroring the output dir to HF."""
state = jax.device_get(jax.tree_util.tree_map(lambda x: x[0], state))
state_dict = {
"params": state.params,
"ema_params1": state.ema_params1,
"opt_state": state.opt_state,
"step": int(state.step),
"epoch": int(state.epoch),
"dropout_rng": state.dropout_rng,
}
ckpt_dir = _local_path(output_dir)
os.makedirs(ckpt_dir, exist_ok=True)
log_for_0(f"Saving checkpoint to {ckpt_dir}")
checkpoints.save_checkpoint_multiprocess(
ckpt_dir, state_dict, step, keep=10, overwrite=True,
)
log_for_0(f"Checkpoint written to {ckpt_dir}")
upload_output_dir_to_hf(output_dir, hf_repo_id, reason="checkpoint")
# ============================================
# Encoder checkpoint (single pickle file)
# ============================================
def load_encoder_checkpoint(checkpoint_path: str):
"""Load a pickled encoder checkpoint from HF first, then local fallback.
HF form: '<org>/<repo>/<filename>'.
"""
if not checkpoint_path:
raise ValueError(
"encoder_checkpoint is not set. Provide a local path or HF Hub path "
"like 'embedded-language-flows/t5_small_encoder_jax/t5_small_encoder_jax.pkl'."
)
log_for_0(f"Loading encoder checkpoint from {checkpoint_path}...")
loaded_params, loaded_from = None, None
errors = []
try:
hf_path = _download_hf_file(checkpoint_path)
if hf_path:
loaded_params = _load_pickle(hf_path)
loaded_from = "HF"
except Exception as e:
errors.append(f"HF: {e}")
log_for_0(f"HF encoder checkpoint load failed ({e}); falling back to local path.")
if loaded_params is None:
local_path = _local_path(checkpoint_path)
try:
loaded_params = _load_pickle(local_path)
loaded_from = "local"
except Exception as e:
errors.append(f"local: {e}")
raise FileNotFoundError(
f"Failed to load encoder checkpoint from {checkpoint_path}. "
f"Tried: {'; '.join(errors)}"
) from e
if isinstance(loaded_params, dict) and "params" in loaded_params:
loaded_params = loaded_params["params"]
log_for_0(f"Loaded {loaded_from} encoder checkpoint.")
return loaded_params
def _load_pickle(path: str):
log_for_0(f"Loading encoder checkpoint from {path}...")
with open(path, "rb") as f:
return pickle.load(f)
def _download_hf_file(path: str) -> Optional[str]:
"""Download a single file from HF Hub and return its local cache path."""
hf_path = _split_hf_path(path, min_parts=3)
if hf_path is None:
return None
repo_id, filename = hf_path
try:
from huggingface_hub import hf_hub_download
log_for_0(f"Downloading checkpoint file from HF: {repo_id}/{filename}")
return hf_hub_download(repo_id=repo_id, filename=filename, repo_type="model")
except Exception as e:
raise FileNotFoundError(f"HF checkpoint file not found: {path} ({e})") from e
def _checkpoint_step(checkpoint_name: str) -> int:
"""Extract the trailing checkpoint step from a name; -1 if absent."""
match = re.search(r"(\d+)$", checkpoint_name)
return int(match.group(1)) if match else -1
# ============================================
# Resume: list + load (local or HF)
# ============================================
def find_all_checkpoints(ckpt_dir: str, prefix: str = "checkpoint_"):
"""Find local checkpoint paths in a directory, sorted by step ascending."""
ckpt_dir = _local_path(ckpt_dir)
if not os.path.isdir(ckpt_dir):
return []
names = sorted(
[f for f in os.listdir(ckpt_dir) if f.startswith(prefix)],
key=_checkpoint_step,
)
return [os.path.join(ckpt_dir, name) for name in names]
def find_latest_checkpoint(ckpt_dir: str, prefix: str = "checkpoint_"):
"""Return the latest local checkpoint path, or None."""
all_ckpts = find_all_checkpoints(ckpt_dir, prefix)
return all_ckpts[-1] if all_ckpts else None
def _download_hf_checkpoint(checkpoint_path: str) -> Optional[str]:
"""Download an HF checkpoint snapshot and return the local checkpoint path."""
hf_path = _split_hf_path(checkpoint_path, min_parts=2)
if hf_path is None:
return None
repo_id, sub_path = hf_path
from huggingface_hub import snapshot_download
log_for_0(f"Downloading checkpoint from HF: {repo_id}" + (f"/{sub_path}" if sub_path else ""))
local_dir = snapshot_download(
repo_id=repo_id, repo_type="model",
allow_patterns=[f"{sub_path}/**"] if sub_path else None,
)
return os.path.join(local_dir, sub_path) if sub_path else local_dir
def _checkpoint_target(state_template: Any):
return {
"params": state_template.params,
"ema_params1": state_template.ema_params1,
"opt_state": state_template.opt_state,
"step": state_template.step,
"epoch": state_template.epoch,
"dropout_rng": state_template.dropout_rng,
}
def _restore_checkpoint(checkpoint_path: str, target: Any):
"""Restore a checkpoint from a file or directory.
Tries (in order):
1. flax.serialization.from_bytes on a file (format written by save_checkpoint)
2. flax.training.checkpoints.restore_checkpoint for HF pre-trained checkpoints
that may have been saved with the old Flax msgpack / orbax format.
"""
local = _local_path(checkpoint_path)
# Resolve directory → latest checkpoint file
resolved = local
if os.path.isdir(local):
latest = find_latest_checkpoint(local)
if latest is not None and os.path.isfile(latest):
resolved = latest
if os.path.isfile(resolved):
try:
with open(resolved, "rb") as f:
data = f.read()
return serialization.from_bytes(target, data)
except Exception:
pass
# Fallback: old Flax/orbax format (e.g., HF pre-trained checkpoints saved before
# this change).
try:
from flax.training import checkpoints as _ckpts
return _ckpts.restore_checkpoint(local, target=target)
except Exception:
return None
def _validate_checkpoint(ckpt: Any):
if ckpt is None:
raise ValueError("checkpoint restore returned None")
required_keys = ("params", "opt_state", "step", "epoch", "dropout_rng")
missing_keys = [key for key in required_keys if key not in ckpt]
if missing_keys:
raise ValueError(f"checkpoint restore missing keys: {missing_keys}")
def load_checkpoint(checkpoint_path: str, state_template: Any) -> Tuple[Any, int]:
"""Load an ELF checkpoint.
Uses an existing local path first; otherwise tries HF and then local fallback.
"""
log_for_0(f"Loading ELF checkpoint from {checkpoint_path}...")
target = _checkpoint_target(state_template)
ckpt, loaded_from = None, None
errors = []
local_path = _local_path(checkpoint_path)
if os.path.exists(local_path):
try:
log_for_0(f"Loading local checkpoint from {local_path}...")
ckpt = _restore_checkpoint(local_path, target)
_validate_checkpoint(ckpt)
loaded_from = "local"
except Exception as e:
errors.append(f"local: {e}")
if ckpt is None:
try:
hf_path = _download_hf_checkpoint(checkpoint_path)
if hf_path:
log_for_0(f"Loading HF checkpoint from {hf_path}...")
ckpt = _restore_checkpoint(hf_path, target)
_validate_checkpoint(ckpt)
loaded_from = "HF"
except Exception as e:
errors.append(f"HF: {e}")
log_for_0(f"HF checkpoint restore failed ({e}); falling back to local path.")
if ckpt is None and not os.path.exists(local_path):
try:
log_for_0(f"Loading local checkpoint from {local_path}...")
ckpt = _restore_checkpoint(local_path, target)
_validate_checkpoint(ckpt)
loaded_from = "local"
except Exception as e:
errors.append(f"local: {e}")
if ckpt is None:
raise ValueError(
f"Failed to load checkpoint from {checkpoint_path}. "
f"Tried: {'; '.join(errors)}"
)
log_for_0(f"Loaded checkpoint keys: {ckpt.keys()}")
restored_state = state_template.replace(
params=jax.tree_util.tree_map(jnp.array, ckpt["params"]),
ema_params1=jax.tree_util.tree_map(jnp.array, ckpt.get("ema_params1", ckpt["params"])),
opt_state=ckpt["opt_state"],
step=ckpt["step"],
epoch=ckpt["epoch"],
dropout_rng=jnp.array(ckpt["dropout_rng"]),
)
step, epoch = int(ckpt["step"]), int(ckpt["epoch"])
log_for_0(f"Loaded {loaded_from} checkpoint from step {step} (epoch {epoch})")
return restored_state, step
|