Function random_dataloader

tests/unit/simple_model.py:275–279 · view source on GitHub ↗

(model, total_samples, hidden_dim, device, dtype=preferred_dtype())

Source from the content-addressed store, hash-verified

273
274
275	def random_dataloader(model, total_samples, hidden_dim, device, dtype=preferred_dtype()):
276	batch_size = model.train_micro_batch_size_per_gpu()
277	train_dataset = random_dataset(total_samples, hidden_dim, device, dtype=dtype)
278	train_loader = torch.utils.data.DataLoader(train_dataset, batch_size=batch_size)
279	return train_loader
280
281
282	def sequence_dataloader(model, total_samples, hidden_dim, device, seq_len: int = 32, dtype=preferred_dtype()):

test_modelMethod · 0.90

testMethod · 0.90

test_flatten_on_accelerator_training_stepMethod · 0.90

test_non_reentrant_single_count_callMethod · 0.90

test_training_step_succeeds_after_fixMethod · 0.90

test_loss_backward_matches_ddpMethod · 0.90

test_grad_accumulationMethod · 0.90

test_grad_accumulation_scale_wrt_gas_falseMethod · 0.90

test_error_when_backward_without_scale_sequentialMethod · 0.90

test_scale_backward_matches_engine_backwardMethod · 0.90

test_scale_backward_matches_ddpMethod · 0.90

test_scale_with_gradient_accumulationMethod · 0.90

preferred_dtypeFunction · 0.85

random_datasetFunction · 0.85

train_micro_batch_size_per_gpuMethod · 0.80

test_modelMethod · 0.72

testMethod · 0.72

test_flatten_on_accelerator_training_stepMethod · 0.72

test_non_reentrant_single_count_callMethod · 0.72

test_training_step_succeeds_after_fixMethod · 0.72

test_loss_backward_matches_ddpMethod · 0.72

test_grad_accumulationMethod · 0.72

test_grad_accumulation_scale_wrt_gas_falseMethod · 0.72

test_error_when_backward_without_scale_sequentialMethod · 0.72

test_scale_backward_matches_engine_backwardMethod · 0.72

test_scale_backward_matches_ddpMethod · 0.72

test_scale_with_gradient_accumulationMethod · 0.72

searching dependent graphs…