You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

221 lines
8.5 KiB

# Ultralytics YOLO 🚀, GPL-3.0 license
import json
import signal
import sys
from pathlib import Path
from time import sleep, time
import requests
from ultralytics.hub.utils import HUB_API_ROOT, check_dataset_disk_space, smart_request
from ultralytics.yolo.utils import LOGGER, PREFIX, __version__, emojis, is_colab, threaded
from ultralytics.yolo.utils.torch_utils import get_flops, get_num_params
AGENT_NAME = f'python-{__version__}-colab' if is_colab() else f'python-{__version__}-local'
session = None
class HubTrainingSession:
def __init__(self, model_id, auth):
self.agent_id = None # identifies which instance is communicating with server
self.model_id = model_id
self.api_url = f'{HUB_API_ROOT}/v1/models/{model_id}'
self.auth_header = auth.get_auth_header()
self._rate_limits = {'metrics': 3.0, 'ckpt': 900.0, 'heartbeat': 300.0} # rate limits (seconds)
self._timers = {} # rate limit timers (seconds)
self._metrics_queue = {} # metrics queue
self.model = self._get_model()
self._start_heartbeat() # start heartbeats
self._register_signal_handlers()
def _register_signal_handlers(self):
signal.signal(signal.SIGTERM, self._handle_signal)
signal.signal(signal.SIGINT, self._handle_signal)
def _handle_signal(self, signum, frame):
"""
Prevent heartbeats from being sent on Colab after kill.
This method does not use frame, it is included as it is
passed by signal.
"""
if self.alive is True:
LOGGER.info(f'{PREFIX}Kill signal received! ❌')
self._stop_heartbeat()
sys.exit(signum)
def _stop_heartbeat(self):
"""End the heartbeat loop"""
self.alive = False
def upload_metrics(self):
payload = {'metrics': self._metrics_queue.copy(), 'type': 'metrics'}
smart_request(f'{self.api_url}', json=payload, headers=self.auth_header, code=2)
def upload_model(self, epoch, weights, is_best=False, map=0.0, final=False):
# Upload a model to HUB
file = None
if Path(weights).is_file():
with open(weights, 'rb') as f:
file = f.read()
if final:
smart_request(
f'{self.api_url}/upload',
data={
'epoch': epoch,
'type': 'final',
'map': map},
files={'best.pt': file},
headers=self.auth_header,
retry=10,
timeout=3600,
code=4,
)
else:
smart_request(
f'{self.api_url}/upload',
data={
'epoch': epoch,
'type': 'epoch',
'isBest': bool(is_best)},
headers=self.auth_header,
files={'last.pt': file},
code=3,
)
def _get_model(self):
# Returns model from database by id
api_url = f'{HUB_API_ROOT}/v1/models/{self.model_id}'
headers = self.auth_header
try:
response = smart_request(api_url, method='get', headers=headers, thread=False, code=0)
data = response.json().get('data', None)
if data.get('status', None) == 'trained':
raise ValueError(
emojis(f'Model is already trained and uploaded to '
f'https://hub.ultralytics.com/models/{self.model_id} 🚀'))
if not data.get('data', None):
raise ValueError('Dataset may still be processing. Please wait a minute and try again.') # RF fix
self.model_id = data['id']
# TODO: restore when server keys when dataset URL and GPU train is working
self.train_args = {
'batch': data['batch_size'],
'epochs': data['epochs'],
'imgsz': data['imgsz'],
'patience': data['patience'],
'device': data['device'],
'cache': data['cache'],
'data': data['data']}
self.input_file = data.get('cfg', data['weights'])
# hack for yolov5 cfg adds u
if 'cfg' in data and 'yolov5' in data['cfg']:
self.input_file = data['cfg'].replace('.yaml', 'u.yaml')
return data
except requests.exceptions.ConnectionError as e:
raise ConnectionRefusedError('ERROR: The HUB server is not online. Please try again later.') from e
except Exception:
raise
def check_disk_space(self):
if not check_dataset_disk_space(self.model['data']):
raise MemoryError('Not enough disk space')
def register_callbacks(self, trainer):
trainer.add_callback('on_pretrain_routine_end', self.on_pretrain_routine_end)
trainer.add_callback('on_fit_epoch_end', self.on_fit_epoch_end)
trainer.add_callback('on_model_save', self.on_model_save)
trainer.add_callback('on_train_end', self.on_train_end)
def on_pretrain_routine_end(self, trainer):
"""
Start timer for upload rate limit.
This method does not use trainer. It is passed to all callbacks by default.
"""
# Start timer for upload rate limit
LOGGER.info(f'{PREFIX}View model at https://hub.ultralytics.com/models/{self.model_id} 🚀')
self._timers = {'metrics': time(), 'ckpt': time()} # start timer on self.rate_limit
def on_fit_epoch_end(self, trainer):
# Upload metrics after val end
all_plots = {**trainer.label_loss_items(trainer.tloss, prefix='train'), **trainer.metrics}
if trainer.epoch == 0:
model_info = {
'model/parameters': get_num_params(trainer.model),
'model/GFLOPs': round(get_flops(trainer.model), 3),
'model/speed(ms)': round(trainer.validator.speed[1], 3)}
all_plots = {**all_plots, **model_info}
self._metrics_queue[trainer.epoch] = json.dumps(all_plots)
if time() - self._timers['metrics'] > self._rate_limits['metrics']:
self.upload_metrics()
self._timers['metrics'] = time() # reset timer
self._metrics_queue = {} # reset queue
def on_model_save(self, trainer):
# Upload checkpoints with rate limiting
is_best = trainer.best_fitness == trainer.fitness
if time() - self._timers['ckpt'] > self._rate_limits['ckpt']:
LOGGER.info(f'{PREFIX}Uploading checkpoint {self.model_id}')
self._upload_model(trainer.epoch, trainer.last, is_best)
self._timers['ckpt'] = time() # reset timer
def on_train_end(self, trainer):
# Upload final model and metrics with exponential standoff
LOGGER.info(f'{PREFIX}Training completed successfully ✅')
LOGGER.info(f'{PREFIX}Uploading final {self.model_id}')
# hack for fetching mAP
mAP = trainer.metrics.get('metrics/mAP50-95(B)', 0)
self._upload_model(trainer.epoch, trainer.best, map=mAP, final=True) # results[3] is mAP0.5:0.95
self.alive = False # stop heartbeats
LOGGER.info(f'{PREFIX}View model at https://hub.ultralytics.com/models/{self.model_id} 🚀')
def _upload_model(self, epoch, weights, is_best=False, map=0.0, final=False):
# Upload a model to HUB
file = None
if Path(weights).is_file():
with open(weights, 'rb') as f:
file = f.read()
file_param = {'best.pt' if final else 'last.pt': file}
endpoint = f'{self.api_url}/upload'
data = {'epoch': epoch}
if final:
data.update({'type': 'final', 'map': map})
else:
data.update({'type': 'epoch', 'isBest': bool(is_best)})
smart_request(
endpoint,
data=data,
files=file_param,
headers=self.auth_header,
retry=10 if final else None,
timeout=3600 if final else None,
code=4 if final else 3,
)
@threaded
def _start_heartbeat(self):
self.alive = True
while self.alive:
r = smart_request(
f'{HUB_API_ROOT}/v1/agent/heartbeat/models/{self.model_id}',
json={
'agent': AGENT_NAME,
'agentId': self.agent_id},
headers=self.auth_header,
retry=0,
code=5,
thread=False,
)
self.agent_id = r.json().get('data', {}).get('agentId', None)
sleep(self._rate_limits['heartbeat'])