events.out.tfevents.1601109183.e8dd0ea41ee1

�K"	�/���A
brain.Event:2�Q
Y.��W�	rQ�/���A
*!

metrics/CrossEntropyLossN<'A)��8J���	z��/���A
*�
�

gin_configB�
B�#### Parameters for Adam:

    Adam.b1 = 0.9
    Adam.b2 = 0.999
    Adam.clip_grad_norm = None
    Adam.eps = 1e-05
    Adam.weight_decay_rate = 1e-05
    
#### Parameters for AddLossWeights:

    # None.
    
#### Parameters for backend:

    backend.name = 'jax'
    
#### Parameters for BucketByLength:

    BucketByLength.length_axis = 0
    BucketByLength.length_keys = None
    BucketByLength.strict_pad_on_len = False
    
#### Parameters for FastGelu:

    # None.
    
#### Parameters for FilterByLength:

    FilterByLength.length_axis = 0
    FilterByLength.length_keys = None
    
#### Parameters for LogSoftmax:

    LogSoftmax.axis = -1
    
#### Parameters for random_spans_helper:

    # None.
    
#### Parameters for layers.SelfAttention:

    layers.SelfAttention.attention_dropout = 0.0
    layers.SelfAttention.bias = False
    layers.SelfAttention.chunk_len = None
    layers.SelfAttention.masked = False
    layers.SelfAttention.n_chunks_after = 0
    layers.SelfAttention.n_chunks_before = 0
    layers.SelfAttention.n_parallel_heads = None
    layers.SelfAttention.predict_drop_len = None
    layers.SelfAttention.predict_mem_len = None
    layers.SelfAttention.share_qk = False
    layers.SelfAttention.use_python_loop = False
    layers.SelfAttention.use_reference_code = False
    
#### Parameters for SentencePieceVocabulary:

    # None.
    
#### Parameters for Serial:

    # None.
    
#### Parameters for Shuffle:

    Shuffle.queue_size = 1024
    
#### Parameters for data.Tokenize:

    # None.
    
#### Parameters for tf_inputs.Tokenize:

    tf_inputs.Tokenize.keys = None
    tf_inputs.Tokenize.n_reserved_ids = 0
    tf_inputs.Tokenize.vocab_type = 'subword'
    
#### Parameters for Vocabulary:

    # None.
    
#### Parameters for warmup_and_rsqrt_decay:

    # None.J

text�vG�,���E	���/���A
*

training/learning_rateĚ'7�-��/m]P	=��/���A
*"
 
training/steps per secondOڄ<2�^p+��K	��/���A
*

training/gradients_l2:�"@���#��wC	���/���A
*


training/lossN<'A�`y�)7�_	T��/���A
*

training/weights_l2�E�Ҽ,.��W�	��[���A
*!

metrics/CrossEntropyLossL�$AB��,���E	5�[���A
*

training/learning_rateu��8;�s/m]P	͢[���A
*"
 
training/steps per secondU�d=
�+��K	��[���A
*

training/gradients_l2H,@;��#��wC	[�[���A
*


training/lossL�$Ay#L�)7�_	�[���A
*

training/weights_l2�E��R