Spaces:
Sleeping
Sleeping
File size: 14,769 Bytes
33d4721 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 |
"""
Common classes and functions for all trainers.
"""
import json
import os
import shutil
import time
import traceback
import requests
from accelerate import PartialState
from huggingface_hub import HfApi
from pydantic import BaseModel
from transformers import TrainerCallback, TrainerControl, TrainerState, TrainingArguments
from autotrain import is_colab, logger
ALLOW_REMOTE_CODE = os.environ.get("ALLOW_REMOTE_CODE", "true").lower() == "true"
def get_file_sizes(directory):
"""
Calculate the sizes of all files in a given directory and its subdirectories.
Args:
directory (str): The path to the directory to scan for files.
Returns:
dict: A dictionary where the keys are the file paths and the values are the file sizes in gigabytes (GB).
"""
file_sizes = {}
for root, _, files in os.walk(directory):
for file in files:
file_path = os.path.join(root, file)
file_size = os.path.getsize(file_path)
file_size_gb = file_size / (1024**3) # Convert bytes to GB
file_sizes[file_path] = file_size_gb
return file_sizes
def remove_global_step(directory):
"""
Removes directories that start with 'global_step' within the specified directory.
This function traverses the given directory and its subdirectories in a bottom-up manner.
If it finds any directory whose name starts with 'global_step', it deletes that directory
and all its contents.
Args:
directory (str): The path to the directory to be traversed and cleaned.
Returns:
None
"""
for root, dirs, _ in os.walk(directory, topdown=False):
for name in dirs:
if name.startswith("global_step"):
folder_path = os.path.join(root, name)
print(f"Removing folder: {folder_path}")
shutil.rmtree(folder_path)
def remove_autotrain_data(config):
"""
Removes the AutoTrain data directory and global step for a given project.
Args:
config (object): Configuration object that contains the project name.
Raises:
OSError: If the removal of the directory fails.
"""
os.system(f"rm -rf {config.project_name}/autotrain-data")
remove_global_step(config.project_name)
def save_training_params(config):
"""
Saves the training parameters to a JSON file, excluding the "token" key if it exists.
Args:
config (object): Configuration object that contains the project name.
The function checks if a file named 'training_params.json' exists in the directory
specified by `config.project_name`. If the file exists, it loads the JSON content,
removes the "token" key if present, and then writes the updated content back to the file.
"""
if os.path.exists(f"{config.project_name}/training_params.json"):
training_params = json.load(open(f"{config.project_name}/training_params.json"))
if "token" in training_params:
training_params.pop("token")
json.dump(
training_params,
open(f"{config.project_name}/training_params.json", "w"),
indent=4,
)
def pause_endpoint(params):
"""
Pauses a Hugging Face endpoint using the provided parameters.
Args:
params (dict or object): Parameters containing the token required for authorization.
If a dictionary is provided, it should have a key "token" with the authorization token.
If an object is provided, it should have an attribute `token` with the authorization token.
Returns:
dict: The JSON response from the API call to pause the endpoint.
Raises:
KeyError: If the "token" key is missing in the params dictionary.
requests.exceptions.RequestException: If there is an issue with the API request.
Environment Variables:
ENDPOINT_ID: Should be set to the endpoint identifier in the format "username/project_name".
"""
if isinstance(params, dict):
token = params["token"]
else:
token = params.token
endpoint_id = os.environ["ENDPOINT_ID"]
username = endpoint_id.split("/")[0]
project_name = endpoint_id.split("/")[1]
api_url = f"https://api.endpoints.huggingface.cloud/v2/endpoint/{username}/{project_name}/pause"
headers = {"Authorization": f"Bearer {token}"}
r = requests.post(api_url, headers=headers, timeout=120)
return r.json()
def pause_space(params, is_failure=False):
"""
Pauses the Hugging Face space and optionally shuts down the endpoint.
This function checks for the presence of "SPACE_ID" and "ENDPOINT_ID" in the environment variables.
If "SPACE_ID" is found, it pauses the space and creates a discussion on the Hugging Face platform
to notify the user about the status of the training run (success or failure).
If "ENDPOINT_ID" is found, it pauses the endpoint.
Args:
params (object): An object containing the necessary parameters, including the token, username, and project name.
is_failure (bool, optional): A flag indicating whether the training run failed. Defaults to False.
Raises:
Exception: If there is an error while creating the discussion on the Hugging Face platform.
Logs:
Info: Logs the status of pausing the space and endpoint.
Warning: Logs any issues encountered while creating the discussion.
Error: Logs if the model failed to train and the discussion was not created.
"""
if "SPACE_ID" in os.environ:
# shut down the space
logger.info("Pausing space...")
api = HfApi(token=params.token)
if is_failure:
msg = "Your training run has failed! Please check the logs for more details"
title = "Your training has failed ❌"
else:
msg = "Your training run was successful! [Check out your trained model here]"
msg += f"(https://huggingface.co/{params.username}/{params.project_name})"
title = "Your training has finished successfully ✅"
if not params.token.startswith("hf_oauth_"):
try:
api.create_discussion(
repo_id=os.environ["SPACE_ID"],
title=title,
description=msg,
repo_type="space",
)
except Exception as e:
logger.warning(f"Failed to create discussion: {e}")
if is_failure:
logger.error("Model failed to train and discussion was not created.")
else:
logger.warning("Model trained successfully but discussion was not created.")
api.pause_space(repo_id=os.environ["SPACE_ID"])
if "ENDPOINT_ID" in os.environ:
# shut down the endpoint
logger.info("Pausing endpoint...")
pause_endpoint(params)
def monitor(func):
"""
A decorator that wraps a function to monitor its execution and handle exceptions.
This decorator performs the following actions:
1. Retrieves the 'config' parameter from the function's keyword arguments or positional arguments.
2. Executes the wrapped function.
3. If an exception occurs during the execution of the wrapped function, logs the error message and stack trace.
4. Optionally pauses the execution if the environment variable 'PAUSE_ON_FAILURE' is set to 1.
Args:
func (callable): The function to be wrapped by the decorator.
Returns:
callable: The wrapped function with monitoring capabilities.
"""
def wrapper(*args, **kwargs):
config = kwargs.get("config", None)
if config is None and len(args) > 0:
config = args[0]
try:
return func(*args, **kwargs)
except Exception as e:
error_message = f"""{func.__name__} has failed due to an exception: {traceback.format_exc()}"""
logger.error(error_message)
logger.error(str(e))
if int(os.environ.get("PAUSE_ON_FAILURE", 1)) == 1:
pause_space(config, is_failure=True)
return wrapper
class AutoTrainParams(BaseModel):
"""
AutoTrainParams is a base class for all AutoTrain parameters.
Attributes:
Config (class): Configuration class for Pydantic model.
protected_namespaces (tuple): Protected namespaces for the model.
Methods:
save(output_dir):
Save parameters to a JSON file in the specified output directory.
__str__():
Return a string representation of the parameters, masking the token if present.
__init__(**data):
Initialize the parameters, check for unused/extra parameters, and warn the user if necessary.
Raises ValueError if project_name is not alphanumeric (with hyphens allowed) or exceeds 50 characters.
"""
class Config:
protected_namespaces = ()
def save(self, output_dir):
"""
Save parameters to a json file.
"""
os.makedirs(output_dir, exist_ok=True)
path = os.path.join(output_dir, "training_params.json")
# save formatted json
with open(path, "w", encoding="utf-8") as f:
f.write(self.model_dump_json(indent=4))
def __str__(self):
"""
String representation of the parameters.
"""
data = self.model_dump()
data["token"] = "*****" if data.get("token") else None
return str(data)
def __init__(self, **data):
"""
Initialize the parameters, check for unused/extra parameters and warn the user.
"""
super().__init__(**data)
if len(self.project_name) > 0:
# make sure project_name is always alphanumeric but can have hyphens. if not, raise ValueError
if not self.project_name.replace("-", "").isalnum():
raise ValueError("project_name must be alphanumeric but can contain hyphens")
# project name cannot be more than 50 characters
if len(self.project_name) > 50:
raise ValueError("project_name cannot be more than 50 characters")
# Parameters not supplied by the user
defaults = set(self.model_fields.keys())
supplied = set(data.keys())
not_supplied = defaults - supplied
if not_supplied and not is_colab:
logger.warning(f"Parameters not supplied by user and set to default: {', '.join(not_supplied)}")
# Parameters that were supplied but not used
# This is a naive implementation. It might catch some internal Pydantic params.
unused = supplied - set(self.model_fields)
if unused:
logger.warning(f"Parameters supplied but not used: {', '.join(unused)}")
class UploadLogs(TrainerCallback):
"""
A callback to upload training logs to the Hugging Face Hub.
Args:
config (object): Configuration object containing necessary parameters.
Attributes:
config (object): Configuration object containing necessary parameters.
api (HfApi or None): Instance of HfApi for interacting with the Hugging Face Hub.
last_upload_time (float): Timestamp of the last upload.
Methods:
on_step_end(args, state, control, **kwargs):
Called at the end of each training step. Uploads logs to the Hugging Face Hub if conditions are met.
"""
def __init__(self, config):
self.config = config
self.api = None
self.last_upload_time = 0
if self.config.push_to_hub:
if PartialState().process_index == 0:
self.api = HfApi(token=config.token)
self.api.create_repo(
repo_id=f"{self.config.username}/{self.config.project_name}", repo_type="model", private=True
)
def on_step_end(self, args: TrainingArguments, state: TrainerState, control: TrainerControl, **kwargs):
if self.config.push_to_hub is False:
return control
if not os.path.exists(os.path.join(self.config.project_name, "runs")):
return control
if (state.global_step + 1) % self.config.logging_steps == 0 and self.config.log == "tensorboard":
if PartialState().process_index == 0:
current_time = time.time()
if current_time - self.last_upload_time >= 600:
try:
self.api.upload_folder(
folder_path=os.path.join(self.config.project_name, "runs"),
repo_id=f"{self.config.username}/{self.config.project_name}",
path_in_repo="runs",
)
except Exception as e:
logger.warning(f"Failed to upload logs: {e}")
logger.warning("Continuing training...")
self.last_upload_time = current_time
return control
class LossLoggingCallback(TrainerCallback):
"""
LossLoggingCallback is a custom callback for logging loss during training.
This callback inherits from `TrainerCallback` and overrides the `on_log` method
to remove the "total_flos" key from the logs and log the remaining information
if the current process is the local process zero.
Methods:
on_log(args, state, control, logs=None, **kwargs):
Called when the logs are updated. Removes the "total_flos" key from the logs
and logs the remaining information if the current process is the local process zero.
Args:
args: The training arguments.
state: The current state of the Trainer.
control: The control object for the Trainer.
logs (dict, optional): The logs dictionary containing the training metrics.
**kwargs: Additional keyword arguments.
"""
def on_log(self, args, state, control, logs=None, **kwargs):
_ = logs.pop("total_flos", None)
if state.is_local_process_zero:
logger.info(logs)
class TrainStartCallback(TrainerCallback):
"""
TrainStartCallback is a custom callback for the Trainer class that logs a message when training begins.
Methods:
on_train_begin(args, state, control, **kwargs):
Logs a message indicating that training is starting.
Args:
args: The training arguments.
state: The current state of the Trainer.
control: The control object for the Trainer.
**kwargs: Additional keyword arguments.
"""
def on_train_begin(self, args, state, control, **kwargs):
logger.info("Starting to train...")
|