MCPcopy
hub / github.com/PaddlePaddle/PaddleRec / dataloader_train_loop

Method dataloader_train_loop

tools/static_gpubox_trainer.py:229–266  ·  view source on GitHub ↗
(self, epoch)

Source from the content-addressed store, hash-verified

227 debug=config.get("runner.dataset_debug"))
228
229 def dataloader_train_loop(self, epoch):
230 logger.info("Epoch: {}, Running DataLoader Begin.".format(epoch))
231 batch_id = 0
232 train_run_cost = 0.0
233 total_examples = 0
234 self.reader.start()
235 while True:
236 try:
237 train_start = time.time()
238 # --------------------------------------------------- #
239 fetch_var = self.exe.run(
240 program=paddle.static.default_main_program(),
241 fetch_list=[var for _, var in self.metrics.items()])
242 # --------------------------------------------------- #
243 train_run_cost += time.time() - train_start
244 total_examples += (self.config.get("runner.train_batch_size"))
245 batch_id += 1
246 print_step = int(config.get("runner.print_interval"))
247 if batch_id % print_step == 0:
248 metrics_string = ""
249 for var_idx, var_name in enumerate(self.metrics):
250 metrics_string += "{}: {}, ".format(var_name,
251 fetch_var[var_idx])
252 profiler_string = ""
253 profiler_string += "avg_batch_cost: {} sec, ".format(
254 format((train_run_cost) / print_step, '.5f'))
255 profiler_string += "avg_samples: {}, ".format(
256 format(total_examples / print_step, '.5f'))
257 profiler_string += "ips: {} {}/sec ".format(
258 format(total_examples / (train_run_cost), '.5f'),
259 self.count_method)
260 logger.info("Epoch: {}, Batch: {}, {} {}".format(
261 epoch, batch_id, metrics_string, profiler_string))
262 train_run_cost = 0.0
263 total_examples = 0
264 except paddle.fluid.core.EOFException:
265 self.reader.reset()
266 break
267
268 def recdataset_train_loop(self, epoch):
269 logger.info("Epoch: {}, Running RecDatast Begin.".format(epoch))

Callers 1

run_workerMethod · 0.95

Calls 3

startMethod · 0.80
runMethod · 0.45
resetMethod · 0.45

Tested by

no test coverage detected