When i try to apply the extract_log method from logix.lightning to my transformer i get the following error.
By inspecting the variables i understood that, even if logix tracks all the layers of the Transformer, and thus write_order_key contains all of them, nested_dict just contains logs from the first block, and it appears that the get_from_nested_dict method, if the key isn't in nested_dict, returns by default the whole dictionary itself; I tried to downgrade lightning to an older version but still doesn't work:
AttributeError Traceback (most recent call last)
in <cell line: 18>()
16 dropout_transf = p, learning_rate = lr, positional_encoding_type = pe, is_causal = is_causal, load_state=True)
17
---> 18 trainerX.extract_log(model, loader_train)
19 result = trainerX.influence(model, loader_test)
14 frames
/usr/local/lib/python3.10/dist-packages/logix/logging/mmap.py in write(save_path, filename, data_buffer, write_order_key, dtype)
61 for key in write_order_key:
62 arr = get_from_nested_dict(nested_dict, key)
---> 63 bytes = arr.nbytes
64 data_shape.append(arr.shape)
65 block_size += reduce(lambda x, y: x * y, arr.shape)
AttributeError: 'collections.defaultdict' object has no attribute 'nbytes'
When i try to apply the extract_log method from logix.lightning to my transformer i get the following error.
By inspecting the variables i understood that, even if logix tracks all the layers of the Transformer, and thus write_order_key contains all of them, nested_dict just contains logs from the first block, and it appears that the get_from_nested_dict method, if the key isn't in nested_dict, returns by default the whole dictionary itself; I tried to downgrade lightning to an older version but still doesn't work:
AttributeError Traceback (most recent call last)
in <cell line: 18>()
16 dropout_transf = p, learning_rate = lr, positional_encoding_type = pe, is_causal = is_causal, load_state=True)
17
---> 18 trainerX.extract_log(model, loader_train)
19 result = trainerX.influence(model, loader_test)
14 frames
/usr/local/lib/python3.10/dist-packages/logix/logging/mmap.py in write(save_path, filename, data_buffer, write_order_key, dtype)
61 for key in write_order_key:
62 arr = get_from_nested_dict(nested_dict, key)
---> 63 bytes = arr.nbytes
64 data_shape.append(arr.shape)
65 block_size += reduce(lambda x, y: x * y, arr.shape)
AttributeError: 'collections.defaultdict' object has no attribute 'nbytes'