('===') try: for tensor_name_0, tensor_name_1 in output_name_dict.items(): testing.assert_allclose(output_0[tensor_name_0], output_1[tensor_name_1], rtol=1, err_msg=f'at {tensor_name_0}, {tensor_name_1}') print("torch_complie does not trigger assertion") except Assertion...
The commas are what create a tuple, by packing the values that are the tuple’s items. On the lefthand side of the = in an assignment statement, you normally use a single target. The target can be a simple identifier (also known as a variable), an indexing (such as alist[i] or...
Browse Library Advanced SearchSign In
🐛 Describe the bug When the model has multiple outputs, changing the sequence of the outputs should not affect their values. However, by swapping output sequence, the output of torch.gt differs for the same input in compiled execution. A...
cpu().numpy() for v in model_out_1] output_1 = dict(zip(output_names_1, model_out_1)) print('===') try: for tensor_name_0, tensor_name_1 in output_name_dict.items(): testing.assert_allclose(output_0[tensor_name_0], output_1[tensor_name_1], rtol=1, err_msg=f'at {te...
numpy() for v in model_out_1] output_1 = dict(zip(output_names_1, model_out_1)) print('===') try: for tensor_name_0, tensor_name_1 in output_name_dict.items(): testing.assert_allclose(output_0[tensor_name_0], output_1[tensor_name_1], rtol=1) print("torch_without_complie...