** beam_size 12
RETURNN starting up, version 20181130.185608--git-d30181f, date/time 2018-12-01-09-58-31 (UTC+0100), pid 22317, cwd /work/smt2/makarov/NMT, Python /usr/bin/python3
RETURNN command line options: ['hmm-factorization/en-de/transformer-hmm', '++load_epoch', '114', '++device', 'gpu', '--task', 'search', '++search_data', 'config:dev', '++beam_size', '12', '++need_data', 'False', '++max_seq_length', '0', '++search_output_file', 'hmm-factorization/en-de/hyp/transformer-hmm', '++batch_size', '2000']
Hostname: cluster-cn-258
TensorFlow: 1.9.0 (v1.9.0-0-g25c197e023) (<site-package> in /u/makarov/.local/lib/python3.5/site-packages/tensorflow)
Setup TF inter and intra global thread pools, num_threads None, session opts {'device_count': {'GPU': 0}, 'log_device_placement': False}.
2018-12-01 09:58:32.562859: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
2018-12-01 09:58:32.978259: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1392] Found device 0 with properties:
name: GeForce GTX 1080 Ti major: 6 minor: 1 memoryClockRate(GHz): 1.582
pciBusID: 0000:02:00.0
totalMemory: 10.92GiB freeMemory: 10.76GiB
2018-12-01 09:58:32.978317: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1471] Adding visible gpu devices: 0
2018-12-01 09:58:32.978337: I tensorflow/core/common_runtime/gpu/gpu_device.cc:952] Device interconnect StreamExecutor with strength 1 edge matrix:
2018-12-01 09:58:32.978348: I tensorflow/core/common_runtime/gpu/gpu_device.cc:958]      0
2018-12-01 09:58:32.978358: I tensorflow/core/common_runtime/gpu/gpu_device.cc:971] 0:   N
CUDA_VISIBLE_DEVICES is set to '0'.
2018-12-01 09:58:33.282635: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1471] Adding visible gpu devices: 0
2018-12-01 09:58:33.828527: I tensorflow/core/common_runtime/gpu/gpu_device.cc:952] Device interconnect StreamExecutor with strength 1 edge matrix:
2018-12-01 09:58:33.828579: I tensorflow/core/common_runtime/gpu/gpu_device.cc:958]      0
2018-12-01 09:58:33.828588: I tensorflow/core/common_runtime/gpu/gpu_device.cc:971] 0:   N
2018-12-01 09:58:33.828955: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1084] Created TensorFlow device (/device:GPU:0 with 10409 MB memory) -> physical GPU (device: 0, name: GeForce GTX 1080 Ti, pci bus id: 0000:02:00.0, compute capability: 6.1)
Collecting TensorFlow device list...
Local devices available to TensorFlow:
  1/2: name: "/device:CPU:0"
       device_type: "CPU"
       memory_limit: 268435456
       locality {
       }
       incarnation: 616944120252845792
  2/2: name: "/device:GPU:0"
       device_type: "GPU"
       memory_limit: 10915220685
       locality {
         bus_id: 1
         links {
         }
       }
       incarnation: 955148772328989222
       physical_device_desc: "device: 0, name: GeForce GTX 1080 Ti, pci bus id: 0000:02:00.0, compute capability: 6.1"
Using gpu device 0: GeForce GTX 1080 Ti
Setup tf.Session with options {'device_count': {'GPU': 1}, 'log_device_placement': False} ...
2018-12-01 09:58:38.902018: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1471] Adding visible gpu devices: 0
2018-12-01 09:58:38.902091: I tensorflow/core/common_runtime/gpu/gpu_device.cc:952] Device interconnect StreamExecutor with strength 1 edge matrix:
2018-12-01 09:58:38.902105: I tensorflow/core/common_runtime/gpu/gpu_device.cc:958]      0
2018-12-01 09:58:38.902115: I tensorflow/core/common_runtime/gpu/gpu_device.cc:971] 0:   N
2018-12-01 09:58:38.902372: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1084] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 10409 MB memory) -> physical GPU (device: 0, name: GeForce GTX 1080 Ti, pci bus id: 0000:02:00.0, compute capability: 6.1)
layer root/'data' output: Data(name='data', shape=(None,), dtype='int32', sparse=True, dim=46300)
layer root/'source_embed_raw' output: Data(name='source_embed_raw_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'source_embed_raw': <tf.Tensor 'source_embed_raw/linear/embedding_lookup:0' shape=(?, ?, 512) dtype=float32>
layer root/'source_embed_weighted' output: Data(name='source_embed_weighted_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'source_embed_weighted': <tf.Tensor 'source_embed_weighted/mul:0' shape=(?, ?, 512) dtype=float32>
layer root/'source_embed_with_pos' output: Data(name='source_embed_with_pos_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'source_embed_with_pos': <tf.Tensor 'source_embed_with_pos/add:0' shape=(?, ?, 512) dtype=float32>
layer root/'source_embed' output: Data(name='source_embed_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'source_embed': <tf.Tensor 'source_embed_with_pos/source_embed_with_pos_identity_with_check_numerics_output/Identity:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_01_self_att_laynorm' output: Data(name='enc_01_self_att_laynorm_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_01_self_att_laynorm': <tf.Tensor 'enc_01_self_att_laynorm/add:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_01_self_att_att' output: Data(name='enc_01_self_att_att_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_01_self_att_att': <tf.Tensor 'enc_01_self_att_att/merge_vdim:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_01_self_att_lin' output: Data(name='enc_01_self_att_lin_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_01_self_att_lin': <tf.Tensor 'enc_01_self_att_lin/linear/dot/Reshape_1:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_01_self_att_drop' output: Data(name='enc_01_self_att_drop_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_01_self_att_drop': <tf.Tensor 'enc_01_self_att_lin/enc_01_self_att_lin_identity_with_check_numerics_output/Identity:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_01_self_att_out' output: Data(name='enc_01_self_att_out_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_01_self_att_out': <tf.Tensor 'enc_01_self_att_out/Add:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_01_ff_laynorm' output: Data(name='enc_01_ff_laynorm_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_01_ff_laynorm': <tf.Tensor 'enc_01_ff_laynorm/add:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_01_ff_conv1' output: Data(name='enc_01_ff_conv1_output', shape=(None, 2048))
debug_add_check_numerics_on_output: add for layer 'enc_01_ff_conv1': <tf.Tensor 'enc_01_ff_conv1/activation/Relu:0' shape=(?, ?, 2048) dtype=float32>
layer root/'enc_01_ff_conv2' output: Data(name='enc_01_ff_conv2_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_01_ff_conv2': <tf.Tensor 'enc_01_ff_conv2/linear/add_bias:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_01_ff_drop' output: Data(name='enc_01_ff_drop_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_01_ff_drop': <tf.Tensor 'enc_01_ff_conv2/enc_01_ff_conv2_identity_with_check_numerics_output/Identity:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_01_ff_out' output: Data(name='enc_01_ff_out_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_01_ff_out': <tf.Tensor 'enc_01_ff_out/Add:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_01' output: Data(name='enc_01_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_01': <tf.Tensor 'enc_01_ff_out/enc_01_ff_out_identity_with_check_numerics_output/Identity:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_02_self_att_laynorm' output: Data(name='enc_02_self_att_laynorm_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_02_self_att_laynorm': <tf.Tensor 'enc_02_self_att_laynorm/add:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_02_self_att_att' output: Data(name='enc_02_self_att_att_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_02_self_att_att': <tf.Tensor 'enc_02_self_att_att/merge_vdim:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_02_self_att_lin' output: Data(name='enc_02_self_att_lin_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_02_self_att_lin': <tf.Tensor 'enc_02_self_att_lin/linear/dot/Reshape_1:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_02_self_att_drop' output: Data(name='enc_02_self_att_drop_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_02_self_att_drop': <tf.Tensor 'enc_02_self_att_lin/enc_02_self_att_lin_identity_with_check_numerics_output/Identity:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_02_self_att_out' output: Data(name='enc_02_self_att_out_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_02_self_att_out': <tf.Tensor 'enc_02_self_att_out/Add:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_02_ff_laynorm' output: Data(name='enc_02_ff_laynorm_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_02_ff_laynorm': <tf.Tensor 'enc_02_ff_laynorm/add:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_02_ff_conv1' output: Data(name='enc_02_ff_conv1_output', shape=(None, 2048))
debug_add_check_numerics_on_output: add for layer 'enc_02_ff_conv1': <tf.Tensor 'enc_02_ff_conv1/activation/Relu:0' shape=(?, ?, 2048) dtype=float32>
layer root/'enc_02_ff_conv2' output: Data(name='enc_02_ff_conv2_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_02_ff_conv2': <tf.Tensor 'enc_02_ff_conv2/linear/add_bias:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_02_ff_drop' output: Data(name='enc_02_ff_drop_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_02_ff_drop': <tf.Tensor 'enc_02_ff_conv2/enc_02_ff_conv2_identity_with_check_numerics_output/Identity:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_02_ff_out' output: Data(name='enc_02_ff_out_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_02_ff_out': <tf.Tensor 'enc_02_ff_out/Add:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_02' output: Data(name='enc_02_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_02': <tf.Tensor 'enc_02_ff_out/enc_02_ff_out_identity_with_check_numerics_output/Identity:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_03_self_att_laynorm' output: Data(name='enc_03_self_att_laynorm_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_03_self_att_laynorm': <tf.Tensor 'enc_03_self_att_laynorm/add:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_03_self_att_att' output: Data(name='enc_03_self_att_att_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_03_self_att_att': <tf.Tensor 'enc_03_self_att_att/merge_vdim:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_03_self_att_lin' output: Data(name='enc_03_self_att_lin_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_03_self_att_lin': <tf.Tensor 'enc_03_self_att_lin/linear/dot/Reshape_1:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_03_self_att_drop' output: Data(name='enc_03_self_att_drop_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_03_self_att_drop': <tf.Tensor 'enc_03_self_att_lin/enc_03_self_att_lin_identity_with_check_numerics_output/Identity:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_03_self_att_out' output: Data(name='enc_03_self_att_out_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_03_self_att_out': <tf.Tensor 'enc_03_self_att_out/Add:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_03_ff_laynorm' output: Data(name='enc_03_ff_laynorm_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_03_ff_laynorm': <tf.Tensor 'enc_03_ff_laynorm/add:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_03_ff_conv1' output: Data(name='enc_03_ff_conv1_output', shape=(None, 2048))
debug_add_check_numerics_on_output: add for layer 'enc_03_ff_conv1': <tf.Tensor 'enc_03_ff_conv1/activation/Relu:0' shape=(?, ?, 2048) dtype=float32>
layer root/'enc_03_ff_conv2' output: Data(name='enc_03_ff_conv2_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_03_ff_conv2': <tf.Tensor 'enc_03_ff_conv2/linear/add_bias:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_03_ff_drop' output: Data(name='enc_03_ff_drop_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_03_ff_drop': <tf.Tensor 'enc_03_ff_conv2/enc_03_ff_conv2_identity_with_check_numerics_output/Identity:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_03_ff_out' output: Data(name='enc_03_ff_out_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_03_ff_out': <tf.Tensor 'enc_03_ff_out/Add:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_03' output: Data(name='enc_03_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_03': <tf.Tensor 'enc_03_ff_out/enc_03_ff_out_identity_with_check_numerics_output/Identity:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_04_self_att_laynorm' output: Data(name='enc_04_self_att_laynorm_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_04_self_att_laynorm': <tf.Tensor 'enc_04_self_att_laynorm/add:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_04_self_att_att' output: Data(name='enc_04_self_att_att_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_04_self_att_att': <tf.Tensor 'enc_04_self_att_att/merge_vdim:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_04_self_att_lin' output: Data(name='enc_04_self_att_lin_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_04_self_att_lin': <tf.Tensor 'enc_04_self_att_lin/linear/dot/Reshape_1:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_04_self_att_drop' output: Data(name='enc_04_self_att_drop_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_04_self_att_drop': <tf.Tensor 'enc_04_self_att_lin/enc_04_self_att_lin_identity_with_check_numerics_output/Identity:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_04_self_att_out' output: Data(name='enc_04_self_att_out_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_04_self_att_out': <tf.Tensor 'enc_04_self_att_out/Add:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_04_ff_laynorm' output: Data(name='enc_04_ff_laynorm_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_04_ff_laynorm': <tf.Tensor 'enc_04_ff_laynorm/add:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_04_ff_conv1' output: Data(name='enc_04_ff_conv1_output', shape=(None, 2048))
debug_add_check_numerics_on_output: add for layer 'enc_04_ff_conv1': <tf.Tensor 'enc_04_ff_conv1/activation/Relu:0' shape=(?, ?, 2048) dtype=float32>
layer root/'enc_04_ff_conv2' output: Data(name='enc_04_ff_conv2_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_04_ff_conv2': <tf.Tensor 'enc_04_ff_conv2/linear/add_bias:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_04_ff_drop' output: Data(name='enc_04_ff_drop_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_04_ff_drop': <tf.Tensor 'enc_04_ff_conv2/enc_04_ff_conv2_identity_with_check_numerics_output/Identity:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_04_ff_out' output: Data(name='enc_04_ff_out_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_04_ff_out': <tf.Tensor 'enc_04_ff_out/Add:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_04' output: Data(name='enc_04_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_04': <tf.Tensor 'enc_04_ff_out/enc_04_ff_out_identity_with_check_numerics_output/Identity:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_05_self_att_laynorm' output: Data(name='enc_05_self_att_laynorm_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_05_self_att_laynorm': <tf.Tensor 'enc_05_self_att_laynorm/add:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_05_self_att_att' output: Data(name='enc_05_self_att_att_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_05_self_att_att': <tf.Tensor 'enc_05_self_att_att/merge_vdim:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_05_self_att_lin' output: Data(name='enc_05_self_att_lin_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_05_self_att_lin': <tf.Tensor 'enc_05_self_att_lin/linear/dot/Reshape_1:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_05_self_att_drop' output: Data(name='enc_05_self_att_drop_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_05_self_att_drop': <tf.Tensor 'enc_05_self_att_lin/enc_05_self_att_lin_identity_with_check_numerics_output/Identity:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_05_self_att_out' output: Data(name='enc_05_self_att_out_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_05_self_att_out': <tf.Tensor 'enc_05_self_att_out/Add:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_05_ff_laynorm' output: Data(name='enc_05_ff_laynorm_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_05_ff_laynorm': <tf.Tensor 'enc_05_ff_laynorm/add:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_05_ff_conv1' output: Data(name='enc_05_ff_conv1_output', shape=(None, 2048))
debug_add_check_numerics_on_output: add for layer 'enc_05_ff_conv1': <tf.Tensor 'enc_05_ff_conv1/activation/Relu:0' shape=(?, ?, 2048) dtype=float32>
layer root/'enc_05_ff_conv2' output: Data(name='enc_05_ff_conv2_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_05_ff_conv2': <tf.Tensor 'enc_05_ff_conv2/linear/add_bias:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_05_ff_drop' output: Data(name='enc_05_ff_drop_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_05_ff_drop': <tf.Tensor 'enc_05_ff_conv2/enc_05_ff_conv2_identity_with_check_numerics_output/Identity:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_05_ff_out' output: Data(name='enc_05_ff_out_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_05_ff_out': <tf.Tensor 'enc_05_ff_out/Add:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_05' output: Data(name='enc_05_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_05': <tf.Tensor 'enc_05_ff_out/enc_05_ff_out_identity_with_check_numerics_output/Identity:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_06_self_att_laynorm' output: Data(name='enc_06_self_att_laynorm_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_06_self_att_laynorm': <tf.Tensor 'enc_06_self_att_laynorm/add:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_06_self_att_att' output: Data(name='enc_06_self_att_att_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_06_self_att_att': <tf.Tensor 'enc_06_self_att_att/merge_vdim:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_06_self_att_lin' output: Data(name='enc_06_self_att_lin_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_06_self_att_lin': <tf.Tensor 'enc_06_self_att_lin/linear/dot/Reshape_1:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_06_self_att_drop' output: Data(name='enc_06_self_att_drop_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_06_self_att_drop': <tf.Tensor 'enc_06_self_att_lin/enc_06_self_att_lin_identity_with_check_numerics_output/Identity:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_06_self_att_out' output: Data(name='enc_06_self_att_out_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_06_self_att_out': <tf.Tensor 'enc_06_self_att_out/Add:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_06_ff_laynorm' output: Data(name='enc_06_ff_laynorm_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_06_ff_laynorm': <tf.Tensor 'enc_06_ff_laynorm/add:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_06_ff_conv1' output: Data(name='enc_06_ff_conv1_output', shape=(None, 2048))
debug_add_check_numerics_on_output: add for layer 'enc_06_ff_conv1': <tf.Tensor 'enc_06_ff_conv1/activation/Relu:0' shape=(?, ?, 2048) dtype=float32>
layer root/'enc_06_ff_conv2' output: Data(name='enc_06_ff_conv2_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_06_ff_conv2': <tf.Tensor 'enc_06_ff_conv2/linear/add_bias:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_06_ff_drop' output: Data(name='enc_06_ff_drop_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_06_ff_drop': <tf.Tensor 'enc_06_ff_conv2/enc_06_ff_conv2_identity_with_check_numerics_output/Identity:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_06_ff_out' output: Data(name='enc_06_ff_out_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_06_ff_out': <tf.Tensor 'enc_06_ff_out/Add:0' shape=(?, ?, 512) dtype=float32>
layer root/'enc_06' output: Data(name='enc_06_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'enc_06': <tf.Tensor 'enc_06_ff_out/enc_06_ff_out_identity_with_check_numerics_output/Identity:0' shape=(?, ?, 512) dtype=float32>
layer root/'encoder' output: Data(name='encoder_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'encoder': <tf.Tensor 'encoder/add:0' shape=(?, ?, 512) dtype=float32>
layer root/'dec_01_att_key0' output: Data(name='dec_01_att_key0_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'dec_01_att_key0': <tf.Tensor 'dec_01_att_key0/linear/dot/Reshape_1:0' shape=(?, ?, 512) dtype=float32>
layer root/'dec_01_att_key' output: Data(name='dec_01_att_key_output', shape=(None, 8, 64))
debug_add_check_numerics_on_output: add for layer 'dec_01_att_key': <tf.Tensor 'dec_01_att_key/Reshape:0' shape=(?, ?, 8, 64) dtype=float32>
layer root/'dec_03_att_key0' output: Data(name='dec_03_att_key0_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'dec_03_att_key0': <tf.Tensor 'dec_03_att_key0/linear/dot/Reshape_1:0' shape=(?, ?, 512) dtype=float32>
layer root/'dec_03_att_key' output: Data(name='dec_03_att_key_output', shape=(None, 8, 64))
debug_add_check_numerics_on_output: add for layer 'dec_03_att_key': <tf.Tensor 'dec_03_att_key/Reshape:0' shape=(?, ?, 8, 64) dtype=float32>
layer root/'dec_04_att_value0' output: Data(name='dec_04_att_value0_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'dec_04_att_value0': <tf.Tensor 'dec_04_att_value0/linear/dot/Reshape_1:0' shape=(?, ?, 512) dtype=float32>
layer root/'dec_04_att_value' output: Data(name='dec_04_att_value_output', shape=(None, 8, 64))
debug_add_check_numerics_on_output: add for layer 'dec_04_att_value': <tf.Tensor 'dec_04_att_value/Reshape:0' shape=(?, ?, 8, 64) dtype=float32>
layer root/'dec_06_att_key0' output: Data(name='dec_06_att_key0_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'dec_06_att_key0': <tf.Tensor 'dec_06_att_key0/linear/dot/Reshape_1:0' shape=(?, ?, 512) dtype=float32>
layer root/'dec_06_att_key' output: Data(name='dec_06_att_key_output', shape=(None, 8, 64))
debug_add_check_numerics_on_output: add for layer 'dec_06_att_key': <tf.Tensor 'dec_06_att_key/Reshape:0' shape=(?, ?, 8, 64) dtype=float32>
layer root/'dec_06_att_value0' output: Data(name='dec_06_att_value0_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'dec_06_att_value0': <tf.Tensor 'dec_06_att_value0/linear/dot/Reshape_1:0' shape=(?, ?, 512) dtype=float32>
layer root/'dec_06_att_value' output: Data(name='dec_06_att_value_output', shape=(None, 8, 64))
debug_add_check_numerics_on_output: add for layer 'dec_06_att_value': <tf.Tensor 'dec_06_att_value/Reshape:0' shape=(?, ?, 8, 64) dtype=float32>
layer root/'dec_03_att_value0' output: Data(name='dec_03_att_value0_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'dec_03_att_value0': <tf.Tensor 'dec_03_att_value0/linear/dot/Reshape_1:0' shape=(?, ?, 512) dtype=float32>
layer root/'dec_03_att_value' output: Data(name='dec_03_att_value_output', shape=(None, 8, 64))
debug_add_check_numerics_on_output: add for layer 'dec_03_att_value': <tf.Tensor 'dec_03_att_value/Reshape:0' shape=(?, ?, 8, 64) dtype=float32>
layer root/'dec_05_att_value0' output: Data(name='dec_05_att_value0_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'dec_05_att_value0': <tf.Tensor 'dec_05_att_value0/linear/dot/Reshape_1:0' shape=(?, ?, 512) dtype=float32>
layer root/'dec_05_att_value' output: Data(name='dec_05_att_value_output', shape=(None, 8, 64))
debug_add_check_numerics_on_output: add for layer 'dec_05_att_value': <tf.Tensor 'dec_05_att_value/Reshape:0' shape=(?, ?, 8, 64) dtype=float32>
layer root/'dec_02_att_key0' output: Data(name='dec_02_att_key0_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'dec_02_att_key0': <tf.Tensor 'dec_02_att_key0/linear/dot/Reshape_1:0' shape=(?, ?, 512) dtype=float32>
layer root/'dec_02_att_key' output: Data(name='dec_02_att_key_output', shape=(None, 8, 64))
debug_add_check_numerics_on_output: add for layer 'dec_02_att_key': <tf.Tensor 'dec_02_att_key/Reshape:0' shape=(?, ?, 8, 64) dtype=float32>
layer root/'dec_01_att_value0' output: Data(name='dec_01_att_value0_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'dec_01_att_value0': <tf.Tensor 'dec_01_att_value0/linear/dot/Reshape_1:0' shape=(?, ?, 512) dtype=float32>
layer root/'dec_01_att_value' output: Data(name='dec_01_att_value_output', shape=(None, 8, 64))
debug_add_check_numerics_on_output: add for layer 'dec_01_att_value': <tf.Tensor 'dec_01_att_value/Reshape:0' shape=(?, ?, 8, 64) dtype=float32>
layer root/'dec_04_att_key0' output: Data(name='dec_04_att_key0_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'dec_04_att_key0': <tf.Tensor 'dec_04_att_key0/linear/dot/Reshape_1:0' shape=(?, ?, 512) dtype=float32>
layer root/'dec_04_att_key' output: Data(name='dec_04_att_key_output', shape=(None, 8, 64))
debug_add_check_numerics_on_output: add for layer 'dec_04_att_key': <tf.Tensor 'dec_04_att_key/Reshape:0' shape=(?, ?, 8, 64) dtype=float32>
layer root/'dec_05_att_key0' output: Data(name='dec_05_att_key0_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'dec_05_att_key0': <tf.Tensor 'dec_05_att_key0/linear/dot/Reshape_1:0' shape=(?, ?, 512) dtype=float32>
layer root/'dec_05_att_key' output: Data(name='dec_05_att_key_output', shape=(None, 8, 64))
debug_add_check_numerics_on_output: add for layer 'dec_05_att_key': <tf.Tensor 'dec_05_att_key/Reshape:0' shape=(?, ?, 8, 64) dtype=float32>
layer root/'dec_02_att_value0' output: Data(name='dec_02_att_value0_output', shape=(None, 512))
debug_add_check_numerics_on_output: add for layer 'dec_02_att_value0': <tf.Tensor 'dec_02_att_value0/linear/dot/Reshape_1:0' shape=(?, ?, 512) dtype=float32>
layer root/'dec_02_att_value' output: Data(name='dec_02_att_value_output', shape=(None, 8, 64))
debug_add_check_numerics_on_output: add for layer 'dec_02_att_value': <tf.Tensor 'dec_02_att_value/Reshape:0' shape=(?, ?, 8, 64) dtype=float32>
layer root/'output' output: Data(name='output_output', shape=(None,), dtype='int32', sparse=True, dim=34908, batch_dim_axis=1, beam_size=12)
Rec layer sub net:
  Input layers moved out of loop: (#: 1)
    encoder_int
  Output layers moved out of loop: (#: 0)
    None
  Layers in loop: (#: 142)
    end
    output
    output_prob
    dec_06_att_weights
    dec_06_att_energy
    dec_06_att_query
    dec_06_att_query0
    dec_06_att_laynorm
    dec_06_self_att_out
    dec_05
    dec_05_ff_out
    dec_05_att_out
    dec_05_att_drop
    dec_05_att_lin
    dec_05_att_att
    dec_05_att0
    dec_05_att_weights_drop
    dec_05_att_weights
    dec_05_att_energy
    dec_05_att_query
    dec_05_att_query0
    dec_05_att_laynorm
    dec_05_self_att_out
    dec_04
    dec_04_ff_out
    dec_04_att_out
    dec_04_att_drop
    dec_04_att_lin
    dec_04_att_att
    dec_04_att0
    dec_04_att_weights_drop
    dec_04_att_weights
    dec_04_att_energy
    dec_04_att_query
    dec_04_att_query0
    dec_04_att_laynorm
    dec_04_self_att_out
    dec_03
    dec_03_ff_out
    dec_03_att_out
    dec_03_att_drop
    dec_03_att_lin
    dec_03_att_att
    dec_03_att0
    dec_03_att_weights_drop
    dec_03_att_weights
    dec_03_att_energy
    dec_03_att_query
    dec_03_att_query0
    dec_03_att_laynorm
    dec_03_self_att_out
    dec_02
    dec_02_ff_out
    dec_02_att_out
    dec_02_att_drop
    dec_02_att_lin
    dec_02_att_att
    dec_02_att0
    dec_02_att_weights_drop
    dec_02_att_weights
    dec_02_att_energy
    dec_02_att_query
    dec_02_att_query0
    dec_02_att_laynorm
    dec_02_self_att_out
    dec_01
    dec_01_ff_out
    dec_01_att_out
    dec_01_att_drop
    dec_01_att_lin
    dec_01_att_att
    dec_01_att0
    dec_01_att_weights_drop
    dec_01_att_weights
    dec_01_att_energy
    dec_01_att_query
    dec_01_att_query0
    dec_01_att_laynorm
    dec_01_self_att_out
    dec_01_self_att_drop
    dec_01_self_att_lin
    dec_01_self_att_att
    dec_01_self_att_laynorm
    target_embed
    target_embed_with_pos
    target_embed_weighted
    target_embed_raw
    dec_01_ff_drop
    dec_01_ff_conv2
    dec_01_ff_conv1
    dec_01_ff_laynorm
    dec_02_self_att_drop
    dec_02_self_att_lin
    dec_02_self_att_att
    dec_02_self_att_laynorm
    dec_02_ff_drop
    dec_02_ff_conv2
    dec_02_ff_conv1
    dec_02_ff_laynorm
    dec_03_self_att_drop
    dec_03_self_att_lin
    dec_03_self_att_att
    dec_03_self_att_laynorm
    dec_03_ff_drop
    dec_03_ff_conv2
    dec_03_ff_conv1
    dec_03_ff_laynorm
    dec_04_self_att_drop
    dec_04_self_att_lin
    dec_04_self_att_att
    dec_04_self_att_laynorm
    dec_04_ff_drop
    dec_04_ff_conv2
    dec_04_ff_conv1
    dec_04_ff_laynorm
    dec_05_self_att_drop
    dec_05_self_att_lin
    dec_05_self_att_att
    dec_05_self_att_laynorm
    dec_05_ff_drop
    dec_05_ff_conv2
    dec_05_ff_conv1
    dec_05_ff_laynorm
    dec_06_self_att_drop
    dec_06_self_att_lin
    dec_06_self_att_att
    dec_06_self_att_laynorm
    decoder_int
    decoder
    dec_06
    dec_06_ff_out
    dec_06_att_out
    dec_06_att_drop
    dec_06_att_lin
    dec_06_att_att
    dec_06_att0
    dec_06_att_weights_drop
    dec_06_ff_drop
    dec_06_ff_conv2
    dec_06_ff_conv1
    dec_06_ff_laynorm
    prev_outputs_int
  Unused layers: (#: 0)
    None
layer root/output:rec-subnet-input/'encoder_int' output: Data(name='encoder_int_output', shape=(None, 1000))
debug_add_check_numerics_on_output: add for layer 'encoder_int': <tf.Tensor 'output/rec/encoder_int/linear/dot/Reshape_1:0' shape=(?, ?, 1000) dtype=float32>
Exception creating layer root/'output' of class RecLayer with opts:
{'max_seq_len': <tf.Tensor 'mul:0' shape=() dtype=int32>,
 'n_out': None,
 'name': 'output',
 'network': <TFNetwork 'root' train=False search>,
 'output': Data(name='output_output', shape=(None,), dtype='int32', sparse=True, dim=34908, batch_dim_axis=1, beam_size=12),
 'sources': [],
 'target': 'classes',
 'unit': {'dec_01': {'class': 'copy', 'from': ['dec_01_ff_out']},
          'dec_01_att0': {'base': 'base:dec_01_att_value',
                          'class': 'generic_attention',
                          'weights': 'dec_01_att_weights_drop'},
          'dec_01_att_att': {'axes': 'static',
                             'class': 'merge_dims',
                             'from': ['dec_01_att0']},
          'dec_01_att_drop': {'class': 'dropout',
                              'dropout': 0.1,
                              'from': ['dec_01_att_lin']},
          'dec_01_att_energy': {'class': 'dot',
                                'from': ['base:dec_01_att_key',
                                         'dec_01_att_query'],
                                'red1': -1,
                                'red2': -1,
                                'var1': 'T',
                                'var2': 'T?'},
          'dec_01_att_laynorm': {'class': 'layer_norm',
                                 'from': ['dec_01_self_att_out']},
          'dec_01_att_lin': {'activation': None,
                             'class': 'linear',
                             'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                     "distribution='uniform', "
                                                     'scale=0.78)',
                             'from': ['dec_01_att_att'],
                             'n_out': 512,
                             'with_bias': False},
          'dec_01_att_out': {'class': 'combine',
                             'from': ['dec_01_self_att_out', 'dec_01_att_drop'],
                             'kind': 'add',
                             'n_out': 512},
          'dec_01_att_query': {'axis': 'F',
                               'class': 'split_dims',
                               'dims': (8, 64),
                               'from': ['dec_01_att_query0']},
          'dec_01_att_query0': {'activation': None,
                                'class': 'linear',
                                'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                        "distribution='uniform', "
                                                        'scale=0.78)',
                                'from': ['dec_01_att_laynorm'],
                                'n_out': 512,
                                'with_bias': False},
          'dec_01_att_weights': {'class': 'softmax_over_spatial',
                                 'energy_factor': 0.125,
                                 'from': ['dec_01_att_energy']},
          'dec_01_att_weights_drop': {'class': 'dropout',
                                      'dropout': 0.1,
                                      'dropout_noise_shape': {'*': None},
                                      'from': ['dec_01_att_weights']},
          'dec_01_ff_conv1': {'activation': 'relu',
                              'class': 'linear',
                              'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                      "distribution='uniform', "
                                                      'scale=0.78)',
                              'from': ['dec_01_ff_laynorm'],
                              'n_out': 2048,
                              'with_bias': True},
          'dec_01_ff_conv2': {'activation': None,
                              'class': 'linear',
                              'dropout': 0.1,
                              'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                      "distribution='uniform', "
                                                      'scale=0.78)',
                              'from': ['dec_01_ff_conv1'],
                              'n_out': 512,
                              'with_bias': True},
          'dec_01_ff_drop': {'class': 'dropout',
                             'dropout': 0.1,
                             'from': ['dec_01_ff_conv2']},
          'dec_01_ff_laynorm': {'class': 'layer_norm',
                                'from': ['dec_01_att_out']},
          'dec_01_ff_out': {'class': 'combine',
                            'from': ['dec_01_att_out', 'dec_01_ff_drop'],
                            'kind': 'add',
                            'n_out': 512},
          'dec_01_self_att_att': {'attention_dropout': 0.1,
                                  'attention_left_only': True,
                                  'class': 'self_attention',
                                  'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                          "distribution='uniform', "
                                                          'scale=0.78)',
                                  'from': ['dec_01_self_att_laynorm'],
                                  'n_out': 512,
                                  'num_heads': 8,
                                  'total_key_dim': 512},
          'dec_01_self_att_drop': {'class': 'dropout',
                                   'dropout': 0.1,
                                   'from': ['dec_01_self_att_lin']},
          'dec_01_self_att_laynorm': {'class': 'layer_norm',
                                      'from': ['target_embed']},
          'dec_01_self_att_lin': {'activation': None,
                                  'class': 'linear',
                                  'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                          "distribution='uniform', "
                                                          'scale=0.78)',
                                  'from': ['dec_01_self_att_att'],
                                  'n_out': 512,
                                  'with_bias': False},
          'dec_01_self_att_out': {'class': 'combine',
                                  'from': ['target_embed',
                                           'dec_01_self_att_drop'],
                                  'kind': 'add',
                                  'n_out': 512},
          'dec_02': {'class': 'copy', 'from': ['dec_02_ff_out']},
          'dec_02_att0': {'base': 'base:dec_02_att_value',
                          'class': 'generic_attention',
                          'weights': 'dec_02_att_weights_drop'},
          'dec_02_att_att': {'axes': 'static',
                             'class': 'merge_dims',
                             'from': ['dec_02_att0']},
          'dec_02_att_drop': {'class': 'dropout',
                              'dropout': 0.1,
                              'from': ['dec_02_att_lin']},
          'dec_02_att_energy': {'class': 'dot',
                                'from': ['base:dec_02_att_key',
                                         'dec_02_att_query'],
                                'red1': -1,
                                'red2': -1,
                                'var1': 'T',
                                'var2': 'T?'},
          'dec_02_att_laynorm': {'class': 'layer_norm',
                                 'from': ['dec_02_self_att_out']},
          'dec_02_att_lin': {'activation': None,
                             'class': 'linear',
                             'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                     "distribution='uniform', "
                                                     'scale=0.78)',
                             'from': ['dec_02_att_att'],
                             'n_out': 512,
                             'with_bias': False},
          'dec_02_att_out': {'class': 'combine',
                             'from': ['dec_02_self_att_out', 'dec_02_att_drop'],
                             'kind': 'add',
                             'n_out': 512},
          'dec_02_att_query': {'axis': 'F',
                               'class': 'split_dims',
                               'dims': (8, 64),
                               'from': ['dec_02_att_query0']},
          'dec_02_att_query0': {'activation': None,
                                'class': 'linear',
                                'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                        "distribution='uniform', "
                                                        'scale=0.78)',
                                'from': ['dec_02_att_laynorm'],
                                'n_out': 512,
                                'with_bias': False},
          'dec_02_att_weights': {'class': 'softmax_over_spatial',
                                 'energy_factor': 0.125,
                                 'from': ['dec_02_att_energy']},
          'dec_02_att_weights_drop': {'class': 'dropout',
                                      'dropout': 0.1,
                                      'dropout_noise_shape': {'*': None},
                                      'from': ['dec_02_att_weights']},
          'dec_02_ff_conv1': {'activation': 'relu',
                              'class': 'linear',
                              'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                      "distribution='uniform', "
                                                      'scale=0.78)',
                              'from': ['dec_02_ff_laynorm'],
                              'n_out': 2048,
                              'with_bias': True},
          'dec_02_ff_conv2': {'activation': None,
                              'class': 'linear',
                              'dropout': 0.1,
                              'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                      "distribution='uniform', "
                                                      'scale=0.78)',
                              'from': ['dec_02_ff_conv1'],
                              'n_out': 512,
                              'with_bias': True},
          'dec_02_ff_drop': {'class': 'dropout',
                             'dropout': 0.1,
                             'from': ['dec_02_ff_conv2']},
          'dec_02_ff_laynorm': {'class': 'layer_norm',
                                'from': ['dec_02_att_out']},
          'dec_02_ff_out': {'class': 'combine',
                            'from': ['dec_02_att_out', 'dec_02_ff_drop'],
                            'kind': 'add',
                            'n_out': 512},
          'dec_02_self_att_att': {'attention_dropout': 0.1,
                                  'attention_left_only': True,
                                  'class': 'self_attention',
                                  'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                          "distribution='uniform', "
                                                          'scale=0.78)',
                                  'from': ['dec_02_self_att_laynorm'],
                                  'n_out': 512,
                                  'num_heads': 8,
                                  'total_key_dim': 512},
          'dec_02_self_att_drop': {'class': 'dropout',
                                   'dropout': 0.1,
                                   'from': ['dec_02_self_att_lin']},
          'dec_02_self_att_laynorm': {'class': 'layer_norm',
                                      'from': ['dec_01']},
          'dec_02_self_att_lin': {'activation': None,
                                  'class': 'linear',
                                  'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                          "distribution='uniform', "
                                                          'scale=0.78)',
                                  'from': ['dec_02_self_att_att'],
                                  'n_out': 512,
                                  'with_bias': False},
          'dec_02_self_att_out': {'class': 'combine',
                                  'from': ['dec_01', 'dec_02_self_att_drop'],
                                  'kind': 'add',
                                  'n_out': 512},
          'dec_03': {'class': 'copy', 'from': ['dec_03_ff_out']},
          'dec_03_att0': {'base': 'base:dec_03_att_value',
                          'class': 'generic_attention',
                          'weights': 'dec_03_att_weights_drop'},
          'dec_03_att_att': {'axes': 'static',
                             'class': 'merge_dims',
                             'from': ['dec_03_att0']},
          'dec_03_att_drop': {'class': 'dropout',
                              'dropout': 0.1,
                              'from': ['dec_03_att_lin']},
          'dec_03_att_energy': {'class': 'dot',
                                'from': ['base:dec_03_att_key',
                                         'dec_03_att_query'],
                                'red1': -1,
                                'red2': -1,
                                'var1': 'T',
                                'var2': 'T?'},
          'dec_03_att_laynorm': {'class': 'layer_norm',
                                 'from': ['dec_03_self_att_out']},
          'dec_03_att_lin': {'activation': None,
                             'class': 'linear',
                             'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                     "distribution='uniform', "
                                                     'scale=0.78)',
                             'from': ['dec_03_att_att'],
                             'n_out': 512,
                             'with_bias': False},
          'dec_03_att_out': {'class': 'combine',
                             'from': ['dec_03_self_att_out', 'dec_03_att_drop'],
                             'kind': 'add',
                             'n_out': 512},
          'dec_03_att_query': {'axis': 'F',
                               'class': 'split_dims',
                               'dims': (8, 64),
                               'from': ['dec_03_att_query0']},
          'dec_03_att_query0': {'activation': None,
                                'class': 'linear',
                                'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                        "distribution='uniform', "
                                                        'scale=0.78)',
                                'from': ['dec_03_att_laynorm'],
                                'n_out': 512,
                                'with_bias': False},
          'dec_03_att_weights': {'class': 'softmax_over_spatial',
                                 'energy_factor': 0.125,
                                 'from': ['dec_03_att_energy']},
          'dec_03_att_weights_drop': {'class': 'dropout',
                                      'dropout': 0.1,
                                      'dropout_noise_shape': {'*': None},
                                      'from': ['dec_03_att_weights']},
          'dec_03_ff_conv1': {'activation': 'relu',
                              'class': 'linear',
                              'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                      "distribution='uniform', "
                                                      'scale=0.78)',
                              'from': ['dec_03_ff_laynorm'],
                              'n_out': 2048,
                              'with_bias': True},
          'dec_03_ff_conv2': {'activation': None,
                              'class': 'linear',
                              'dropout': 0.1,
                              'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                      "distribution='uniform', "
                                                      'scale=0.78)',
                              'from': ['dec_03_ff_conv1'],
                              'n_out': 512,
                              'with_bias': True},
          'dec_03_ff_drop': {'class': 'dropout',
                             'dropout': 0.1,
                             'from': ['dec_03_ff_conv2']},
          'dec_03_ff_laynorm': {'class': 'layer_norm',
                                'from': ['dec_03_att_out']},
          'dec_03_ff_out': {'class': 'combine',
                            'from': ['dec_03_att_out', 'dec_03_ff_drop'],
                            'kind': 'add',
                            'n_out': 512},
          'dec_03_self_att_att': {'attention_dropout': 0.1,
                                  'attention_left_only': True,
                                  'class': 'self_attention',
                                  'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                          "distribution='uniform', "
                                                          'scale=0.78)',
                                  'from': ['dec_03_self_att_laynorm'],
                                  'n_out': 512,
                                  'num_heads': 8,
                                  'total_key_dim': 512},
          'dec_03_self_att_drop': {'class': 'dropout',
                                   'dropout': 0.1,
                                   'from': ['dec_03_self_att_lin']},
          'dec_03_self_att_laynorm': {'class': 'layer_norm',
                                      'from': ['dec_02']},
          'dec_03_self_att_lin': {'activation': None,
                                  'class': 'linear',
                                  'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                          "distribution='uniform', "
                                                          'scale=0.78)',
                                  'from': ['dec_03_self_att_att'],
                                  'n_out': 512,
                                  'with_bias': False},
          'dec_03_self_att_out': {'class': 'combine',
                                  'from': ['dec_02', 'dec_03_self_att_drop'],
                                  'kind': 'add',
                                  'n_out': 512},
          'dec_04': {'class': 'copy', 'from': ['dec_04_ff_out']},
          'dec_04_att0': {'base': 'base:dec_04_att_value',
                          'class': 'generic_attention',
                          'weights': 'dec_04_att_weights_drop'},
          'dec_04_att_att': {'axes': 'static',
                             'class': 'merge_dims',
                             'from': ['dec_04_att0']},
          'dec_04_att_drop': {'class': 'dropout',
                              'dropout': 0.1,
                              'from': ['dec_04_att_lin']},
          'dec_04_att_energy': {'class': 'dot',
                                'from': ['base:dec_04_att_key',
                                         'dec_04_att_query'],
                                'red1': -1,
                                'red2': -1,
                                'var1': 'T',
                                'var2': 'T?'},
          'dec_04_att_laynorm': {'class': 'layer_norm',
                                 'from': ['dec_04_self_att_out']},
          'dec_04_att_lin': {'activation': None,
                             'class': 'linear',
                             'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                     "distribution='uniform', "
                                                     'scale=0.78)',
                             'from': ['dec_04_att_att'],
                             'n_out': 512,
                             'with_bias': False},
          'dec_04_att_out': {'class': 'combine',
                             'from': ['dec_04_self_att_out', 'dec_04_att_drop'],
                             'kind': 'add',
                             'n_out': 512},
          'dec_04_att_query': {'axis': 'F',
                               'class': 'split_dims',
                               'dims': (8, 64),
                               'from': ['dec_04_att_query0']},
          'dec_04_att_query0': {'activation': None,
                                'class': 'linear',
                                'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                        "distribution='uniform', "
                                                        'scale=0.78)',
                                'from': ['dec_04_att_laynorm'],
                                'n_out': 512,
                                'with_bias': False},
          'dec_04_att_weights': {'class': 'softmax_over_spatial',
                                 'energy_factor': 0.125,
                                 'from': ['dec_04_att_energy']},
          'dec_04_att_weights_drop': {'class': 'dropout',
                                      'dropout': 0.1,
                                      'dropout_noise_shape': {'*': None},
                                      'from': ['dec_04_att_weights']},
          'dec_04_ff_conv1': {'activation': 'relu',
                              'class': 'linear',
                              'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                      "distribution='uniform', "
                                                      'scale=0.78)',
                              'from': ['dec_04_ff_laynorm'],
                              'n_out': 2048,
                              'with_bias': True},
          'dec_04_ff_conv2': {'activation': None,
                              'class': 'linear',
                              'dropout': 0.1,
                              'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                      "distribution='uniform', "
                                                      'scale=0.78)',
                              'from': ['dec_04_ff_conv1'],
                              'n_out': 512,
                              'with_bias': True},
          'dec_04_ff_drop': {'class': 'dropout',
                             'dropout': 0.1,
                             'from': ['dec_04_ff_conv2']},
          'dec_04_ff_laynorm': {'class': 'layer_norm',
                                'from': ['dec_04_att_out']},
          'dec_04_ff_out': {'class': 'combine',
                            'from': ['dec_04_att_out', 'dec_04_ff_drop'],
                            'kind': 'add',
                            'n_out': 512},
          'dec_04_self_att_att': {'attention_dropout': 0.1,
                                  'attention_left_only': True,
                                  'class': 'self_attention',
                                  'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                          "distribution='uniform', "
                                                          'scale=0.78)',
                                  'from': ['dec_04_self_att_laynorm'],
                                  'n_out': 512,
                                  'num_heads': 8,
                                  'total_key_dim': 512},
          'dec_04_self_att_drop': {'class': 'dropout',
                                   'dropout': 0.1,
                                   'from': ['dec_04_self_att_lin']},
          'dec_04_self_att_laynorm': {'class': 'layer_norm',
                                      'from': ['dec_03']},
          'dec_04_self_att_lin': {'activation': None,
                                  'class': 'linear',
                                  'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                          "distribution='uniform', "
                                                          'scale=0.78)',
                                  'from': ['dec_04_self_att_att'],
                                  'n_out': 512,
                                  'with_bias': False},
          'dec_04_self_att_out': {'class': 'combine',
                                  'from': ['dec_03', 'dec_04_self_att_drop'],
                                  'kind': 'add',
                                  'n_out': 512},
          'dec_05': {'class': 'copy', 'from': ['dec_05_ff_out']},
          'dec_05_att0': {'base': 'base:dec_05_att_value',
                          'class': 'generic_attention',
                          'weights': 'dec_05_att_weights_drop'},
          'dec_05_att_att': {'axes': 'static',
                             'class': 'merge_dims',
                             'from': ['dec_05_att0']},
          'dec_05_att_drop': {'class': 'dropout',
                              'dropout': 0.1,
                              'from': ['dec_05_att_lin']},
          'dec_05_att_energy': {'class': 'dot',
                                'from': ['base:dec_05_att_key',
                                         'dec_05_att_query'],
                                'red1': -1,
                                'red2': -1,
                                'var1': 'T',
                                'var2': 'T?'},
          'dec_05_att_laynorm': {'class': 'layer_norm',
                                 'from': ['dec_05_self_att_out']},
          'dec_05_att_lin': {'activation': None,
                             'class': 'linear',
                             'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                     "distribution='uniform', "
                                                     'scale=0.78)',
                             'from': ['dec_05_att_att'],
                             'n_out': 512,
                             'with_bias': False},
          'dec_05_att_out': {'class': 'combine',
                             'from': ['dec_05_self_att_out', 'dec_05_att_drop'],
                             'kind': 'add',
                             'n_out': 512},
          'dec_05_att_query': {'axis': 'F',
                               'class': 'split_dims',
                               'dims': (8, 64),
                               'from': ['dec_05_att_query0']},
          'dec_05_att_query0': {'activation': None,
                                'class': 'linear',
                                'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                        "distribution='uniform', "
                                                        'scale=0.78)',
                                'from': ['dec_05_att_laynorm'],
                                'n_out': 512,
                                'with_bias': False},
          'dec_05_att_weights': {'class': 'softmax_over_spatial',
                                 'energy_factor': 0.125,
                                 'from': ['dec_05_att_energy']},
          'dec_05_att_weights_drop': {'class': 'dropout',
                                      'dropout': 0.1,
                                      'dropout_noise_shape': {'*': None},
                                      'from': ['dec_05_att_weights']},
          'dec_05_ff_conv1': {'activation': 'relu',
                              'class': 'linear',
                              'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                      "distribution='uniform', "
                                                      'scale=0.78)',
                              'from': ['dec_05_ff_laynorm'],
                              'n_out': 2048,
                              'with_bias': True},
          'dec_05_ff_conv2': {'activation': None,
                              'class': 'linear',
                              'dropout': 0.1,
                              'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                      "distribution='uniform', "
                                                      'scale=0.78)',
                              'from': ['dec_05_ff_conv1'],
                              'n_out': 512,
                              'with_bias': True},
          'dec_05_ff_drop': {'class': 'dropout',
                             'dropout': 0.1,
                             'from': ['dec_05_ff_conv2']},
          'dec_05_ff_laynorm': {'class': 'layer_norm',
                                'from': ['dec_05_att_out']},
          'dec_05_ff_out': {'class': 'combine',
                            'from': ['dec_05_att_out', 'dec_05_ff_drop'],
                            'kind': 'add',
                            'n_out': 512},
          'dec_05_self_att_att': {'attention_dropout': 0.1,
                                  'attention_left_only': True,
                                  'class': 'self_attention',
                                  'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                          "distribution='uniform', "
                                                          'scale=0.78)',
                                  'from': ['dec_05_self_att_laynorm'],
                                  'n_out': 512,
                                  'num_heads': 8,
                                  'total_key_dim': 512},
          'dec_05_self_att_drop': {'class': 'dropout',
                                   'dropout': 0.1,
                                   'from': ['dec_05_self_att_lin']},
          'dec_05_self_att_laynorm': {'class': 'layer_norm',
                                      'from': ['dec_04']},
          'dec_05_self_att_lin': {'activation': None,
                                  'class': 'linear',
                                  'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                          "distribution='uniform', "
                                                          'scale=0.78)',
                                  'from': ['dec_05_self_att_att'],
                                  'n_out': 512,
                                  'with_bias': False},
          'dec_05_self_att_out': {'class': 'combine',
                                  'from': ['dec_04', 'dec_05_self_att_drop'],
                                  'kind': 'add',
                                  'n_out': 512},
          'dec_06': {'class': 'copy', 'from': ['dec_06_ff_out']},
          'dec_06_att0': {'base': 'base:dec_06_att_value',
                          'class': 'generic_attention',
                          'weights': 'dec_06_att_weights_drop'},
          'dec_06_att_att': {'axes': 'static',
                             'class': 'merge_dims',
                             'from': ['dec_06_att0']},
          'dec_06_att_drop': {'class': 'dropout',
                              'dropout': 0.1,
                              'from': ['dec_06_att_lin']},
          'dec_06_att_energy': {'class': 'dot',
                                'from': ['base:dec_06_att_key',
                                         'dec_06_att_query'],
                                'red1': -1,
                                'red2': -1,
                                'var1': 'T',
                                'var2': 'T?'},
          'dec_06_att_laynorm': {'class': 'layer_norm',
                                 'from': ['dec_06_self_att_out']},
          'dec_06_att_lin': {'activation': None,
                             'class': 'linear',
                             'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                     "distribution='uniform', "
                                                     'scale=0.78)',
                             'from': ['dec_06_att_att'],
                             'n_out': 512,
                             'with_bias': False},
          'dec_06_att_out': {'class': 'combine',
                             'from': ['dec_06_self_att_out', 'dec_06_att_drop'],
                             'kind': 'add',
                             'n_out': 512},
          'dec_06_att_query': {'axis': 'F',
                               'class': 'split_dims',
                               'dims': (8, 64),
                               'from': ['dec_06_att_query0']},
          'dec_06_att_query0': {'activation': None,
                                'class': 'linear',
                                'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                        "distribution='uniform', "
                                                        'scale=0.78)',
                                'from': ['dec_06_att_laynorm'],
                                'n_out': 512,
                                'with_bias': False},
          'dec_06_att_weights': {'class': 'softmax_over_spatial',
                                 'energy_factor': 0.125,
                                 'from': ['dec_06_att_energy']},
          'dec_06_att_weights_drop': {'class': 'dropout',
                                      'dropout': 0.1,
                                      'dropout_noise_shape': {'*': None},
                                      'from': ['dec_06_att_weights']},
          'dec_06_ff_conv1': {'activation': 'relu',
                              'class': 'linear',
                              'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                      "distribution='uniform', "
                                                      'scale=0.78)',
                              'from': ['dec_06_ff_laynorm'],
                              'n_out': 2048,
                              'with_bias': True},
          'dec_06_ff_conv2': {'activation': None,
                              'class': 'linear',
                              'dropout': 0.1,
                              'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                      "distribution='uniform', "
                                                      'scale=0.78)',
                              'from': ['dec_06_ff_conv1'],
                              'n_out': 512,
                              'with_bias': True},
          'dec_06_ff_drop': {'class': 'dropout',
                             'dropout': 0.1,
                             'from': ['dec_06_ff_conv2']},
          'dec_06_ff_laynorm': {'class': 'layer_norm',
                                'from': ['dec_06_att_out']},
          'dec_06_ff_out': {'class': 'combine',
                            'from': ['dec_06_att_out', 'dec_06_ff_drop'],
                            'kind': 'add',
                            'n_out': 512},
          'dec_06_self_att_att': {'attention_dropout': 0.1,
                                  'attention_left_only': True,
                                  'class': 'self_attention',
                                  'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                          "distribution='uniform', "
                                                          'scale=0.78)',
                                  'from': ['dec_06_self_att_laynorm'],
                                  'n_out': 512,
                                  'num_heads': 8,
                                  'total_key_dim': 512},
          'dec_06_self_att_drop': {'class': 'dropout',
                                   'dropout': 0.1,
                                   'from': ['dec_06_self_att_lin']},
          'dec_06_self_att_laynorm': {'class': 'layer_norm',
                                      'from': ['dec_05']},
          'dec_06_self_att_lin': {'activation': None,
                                  'class': 'linear',
                                  'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                          "distribution='uniform', "
                                                          'scale=0.78)',
                                  'from': ['dec_06_self_att_att'],
                                  'n_out': 512,
                                  'with_bias': False},
          'dec_06_self_att_out': {'class': 'combine',
                                  'from': ['dec_05', 'dec_06_self_att_drop'],
                                  'kind': 'add',
                                  'n_out': 512},
          'decoder': {'class': 'layer_norm', 'from': ['dec_06'], 'n_out': 512},
          'decoder_int': {'activation': None,
                          'class': 'linear',
                          'from': ['decoder'],
                          'n_out': 1000,
                          'with_bias': False},
          'encoder_int': {'activation': None,
                          'class': 'linear',
                          'from': ['base:encoder'],
                          'n_out': 1000,
                          'with_bias': False},
          'end': {'class': 'compare', 'from': ['output'], 'value': 0},
          'output': {'beam_size': 12,
                     'class': 'choice',
                     'from': ['output_prob'],
                     'initial_output': 0,
                     'target': 'classes'},
          'output_prob': {'attention_weights': 'dec_06_att_weights',
                          'base_encoder_transformed': 'encoder_int',
                          'class': 'hmm_factorization',
                          'debug': False,
                          'from': 'dec_06_att_weights',
                          'loss': 'ce',
                          'n_out': 34908,
                          'prev_outputs': 'prev_outputs_int',
                          'prev_state': 'decoder_int',
                          'target': 'classes',
                          'threshold': None,
                          'transpose_and_average_att_weights': True},
          'prev_outputs_int': {'activation': None,
                               'class': 'linear',
                               'from': ['prev:target_embed_raw'],
                               'n_out': 1000,
                               'with_bias': False},
          'target_embed': {'class': 'dropout',
                           'dropout': 0.0,
                           'from': ['target_embed_with_pos']},
          'target_embed_raw': {'activation': None,
                               'class': 'linear',
                               'forward_weights_init': "variance_scaling_initializer(mode='fan_in', "
                                                       "distribution='uniform', "
                                                       'scale=0.78)',
                               'from': ['prev:output'],
                               'n_out': 512,
                               'with_bias': False},
          'target_embed_weighted': {'class': 'eval',
                                    'eval': 'source(0) * 22.627417',
                                    'from': ['target_embed_raw']},
          'target_embed_with_pos': {'add_to_input': True,
                                    'class': 'positional_encoding',
                                    'from': ['target_embed_weighted']}}}
EXCEPTION
Traceback (most recent call last):
  File "/u/makarov/returnn-hmm-fac/rnn.py", line 591, in <module>
    line: main(sys.argv)
    locals:
      main = <local> <function main at 0x7fe7b0180bf8>
      sys = <local> <module 'sys' (built-in)>
      sys.argv = <local> ['/u/makarov/returnn-hmm-fac/rnn.py', 'hmm-factorization/en-de/transformer-hmm', '++load_epoch', '114', '++device', 'gpu', '--task', 'search', '++search_data', 'config:dev', '++beam_size', '12', '++need_data', 'False', '++max_seq_length', '0', '++search_output_file', 'hmm-factorization/en-de/hyp/..., len = 20, _[0]: {len = 33}
  File "/u/makarov/returnn-hmm-fac/rnn.py", line 579, in main
    line: executeMainTask()
    locals:
      executeMainTask = <global> <function executeMainTask at 0x7fe7b0180ae8>
  File "/u/makarov/returnn-hmm-fac/rnn.py", line 434, in executeMainTask
    line: engine.init_network_from_config(config)
    locals:
      engine = <global> <TFEngine.Engine object at 0x7fe8124a5e48>
      engine.init_network_from_config = <global> <bound method Engine.init_network_from_config of <TFEngine.Engine object at 0x7fe8124a5e48>>
      config = <global> <Config.Config object at 0x7fe810aca080>
  File "/u/makarov/returnn-hmm-fac/TFEngine.py", line 936, in init_network_from_config
    line: self._init_network(net_desc=net_dict, epoch=self.epoch)
    locals:
      self = <local> <TFEngine.Engine object at 0x7fe8124a5e48>
      self._init_network = <local> <bound method Engine._init_network of <TFEngine.Engine object at 0x7fe8124a5e48>>
      net_desc = <not found>
      net_dict = <local> {'enc_05_self_att_att': {'total_key_dim': 512, 'from': ['enc_05_self_att_laynorm'], 'class': 'self_attention', 'forward_weights_init': "variance_scaling_initializer(mode='fan_in', distribution='uniform', scale=0.78)", 'num_heads': 8, 'attention_left_only': False, 'attention_dropout': 0.1, 'n_out'..., len = 97
      epoch = <local> 114
      self.epoch = <local> 114
  File "/u/makarov/returnn-hmm-fac/TFEngine.py", line 1059, in _init_network
    line: self.network, self.updater = self.create_network(
            config=self.config,
            rnd_seed=net_random_seed,
            train_flag=train_flag, eval_flag=self.use_eval_flag, search_flag=self.use_search_flag,
            initial_learning_rate=getattr(self, "initial_learning_rate", None),
            net_dict=net_desc)
    locals:
      self = <local> <TFEngine.Engine object at 0x7fe8124a5e48>
      self.network = <local> None
      self.updater = <local> None
      self.create_network = <local> <bound method Engine.create_network of <class 'TFEngine.Engine'>>
      config = <not found>
      self.config = <local> <Config.Config object at 0x7fe810aca080>
      rnd_seed = <not found>
      net_random_seed = <local> 114
      train_flag = <local> False
      eval_flag = <not found>
      self.use_eval_flag = <local> True
      search_flag = <not found>
      self.use_search_flag = <local> True
      initial_learning_rate = <not found>
      getattr = <builtin> <built-in function getattr>
      net_dict = <not found>
      net_desc = <local> {'enc_05_self_att_att': {'total_key_dim': 512, 'from': ['enc_05_self_att_laynorm'], 'class': 'self_attention', 'forward_weights_init': "variance_scaling_initializer(mode='fan_in', distribution='uniform', scale=0.78)", 'num_heads': 8, 'attention_left_only': False, 'attention_dropout': 0.1, 'n_out'..., len = 97
  File "/u/makarov/returnn-hmm-fac/TFEngine.py", line 1090, in create_network
    line: network.construct_from_dict(net_dict)
    locals:
      network = <local> <TFNetwork 'root' train=False search>
      network.construct_from_dict = <local> <bound method TFNetwork.construct_from_dict of <TFNetwork 'root' train=False search>>
      net_dict = <local> {'enc_05_self_att_att': {'total_key_dim': 512, 'from': ['enc_05_self_att_laynorm'], 'class': 'self_attention', 'forward_weights_init': "variance_scaling_initializer(mode='fan_in', distribution='uniform', scale=0.78)", 'num_heads': 8, 'attention_left_only': False, 'attention_dropout': 0.1, 'n_out'..., len = 97
  File "/u/makarov/returnn-hmm-fac/TFNetwork.py", line 338, in construct_from_dict
    line: self.construct_layer(net_dict, name)
    locals:
      self = <local> <TFNetwork 'root' train=False search>
      self.construct_layer = <local> <bound method TFNetwork.construct_layer of <TFNetwork 'root' train=False search>>
      net_dict = <local> {'enc_05_self_att_att': {'total_key_dim': 512, 'from': ['enc_05_self_att_laynorm'], 'class': 'self_attention', 'forward_weights_init': "variance_scaling_initializer(mode='fan_in', distribution='uniform', scale=0.78)", 'num_heads': 8, 'attention_left_only': False, 'attention_dropout': 0.1, 'n_out'..., len = 97
      name = <local> 'decision', len = 8
  File "/u/makarov/returnn-hmm-fac/TFNetwork.py", line 407, in construct_layer
    line: layer_class.transform_config_dict(layer_desc, network=self, get_layer=get_layer)
    locals:
      layer_class = <local> <class 'TFNetworkRecLayer.DecideLayer'>
      layer_class.transform_config_dict = <local> <bound method LayerBase.transform_config_dict of <class 'TFNetworkRecLayer.DecideLayer'>>
      layer_desc = <local> {'loss_opts': {}, 'target': 'classes', 'loss': 'edit_distance'}
      network = <not found>
      self = <local> <TFNetwork 'root' train=False search>
      get_layer = <local> <function TFNetwork.construct_layer.<locals>.get_layer at 0x7fe7603061e0>
  File "/u/makarov/returnn-hmm-fac/TFNetworkLayer.py", line 358, in transform_config_dict
    line: for src_name in src_names
    locals:
      src_name = <not found>
      src_names = <local> ['output'], _[0]: {len = 6}
  File "/u/makarov/returnn-hmm-fac/TFNetworkLayer.py", line 359, in <listcomp>
    line: d["sources"] = [
            get_layer(src_name)
            for src_name in src_names
            if not src_name == "none"]
    locals:
      d = <not found>
      get_layer = <local> <function TFNetwork.construct_layer.<locals>.get_layer at 0x7fe7603061e0>
      src_name = <local> 'output', len = 6
      src_names = <not found>
  File "/u/makarov/returnn-hmm-fac/TFNetwork.py", line 397, in get_layer
    line: return self.construct_layer(net_dict=net_dict, name=src_name)
    locals:
      self = <local> <TFNetwork 'root' train=False search>
      self.construct_layer = <local> <bound method TFNetwork.construct_layer of <TFNetwork 'root' train=False search>>
      net_dict = <local> {'enc_05_self_att_att': {'total_key_dim': 512, 'from': ['enc_05_self_att_laynorm'], 'class': 'self_attention', 'forward_weights_init': "variance_scaling_initializer(mode='fan_in', distribution='uniform', scale=0.78)", 'num_heads': 8, 'attention_left_only': False, 'attention_dropout': 0.1, 'n_out'..., len = 97
      name = <not found>
      src_name = <local> 'output', len = 6
  File "/u/makarov/returnn-hmm-fac/TFNetwork.py", line 410, in construct_layer
    line: return add_layer(name=name, layer_class=layer_class, **layer_desc)
    locals:
      add_layer = <local> <bound method TFNetwork.add_layer of <TFNetwork 'root' train=False search>>
      name = <local> 'output', len = 6
      layer_class = <local> <class 'TFNetworkRecLayer.RecLayer'>
      layer_desc = <local> {'max_seq_len': <tf.Tensor 'mul:0' shape=() dtype=int32>, 'unit': {'dec_06_att_out': {'from': ['dec_06_self_att_out', 'dec_06_att_drop'], 'kind': 'add', 'class': 'combine', 'n_out': 512}, 'dec_05_att_weights_drop': {'dropout_noise_shape': {'*': None}, 'from': ['dec_05_att_weights'], 'class': 'dro...
  File "/u/makarov/returnn-hmm-fac/TFNetwork.py", line 497, in add_layer
    line: layer = self._create_layer(name=name, layer_class=layer_class, **layer_desc)
    locals:
      layer = <not found>
      self = <local> <TFNetwork 'root' train=False search>
      self._create_layer = <local> <bound method TFNetwork._create_layer of <TFNetwork 'root' train=False search>>
      name = <local> 'output', len = 6
      layer_class = <local> <class 'TFNetworkRecLayer.RecLayer'>
      layer_desc = <local> {'unit': {'dec_06_att_out': {'from': ['dec_06_self_att_out', 'dec_06_att_drop'], 'kind': 'add', 'class': 'combine', 'n_out': 512}, 'dec_05_att_weights_drop': {'dropout_noise_shape': {'*': None}, 'from': ['dec_05_att_weights'], 'class': 'dropout', 'dropout': 0.1}, 'dec_01_att_energy': {'from': ['b...
  File "/u/makarov/returnn-hmm-fac/TFNetwork.py", line 456, in _create_layer
    line: layer = layer_class(**layer_desc)
    locals:
      layer = <not found>
      layer_class = <local> <class 'TFNetworkRecLayer.RecLayer'>
      layer_desc = <local> {'max_seq_len': <tf.Tensor 'mul:0' shape=() dtype=int32>, 'network': <TFNetwork 'root' train=False search>, 'name': 'output', 'unit': {'dec_06_att_out': {'from': ['dec_06_self_att_out', 'dec_06_att_drop'], 'kind': 'add', 'class': 'combine', 'n_out': 512}, 'dec_05_att_weights_drop': {'dropout_nois..., len = 8
  File "/u/makarov/returnn-hmm-fac/TFNetworkRecLayer.py", line 179, in __init__
    line: y = self._get_output_subnet_unit(self.cell)
    locals:
      y = <not found>
      self = <local> <RecLayer 'output' out_type=Data(shape=(None,), dtype='int32', sparse=True, dim=34908, batch_dim_axis=1, beam_size=12)>
      self._get_output_subnet_unit = <local> <bound method RecLayer._get_output_subnet_unit of <RecLayer 'output' out_type=Data(shape=(None,), dtype='int32', sparse=True, dim=34908, batch_dim_axis=1, beam_size=12)>>
      self.cell = <local> <TFNetworkRecLayer._SubnetworkRecCell object at 0x7fe37643e7b8>
  File "/u/makarov/returnn-hmm-fac/TFNetworkRecLayer.py", line 703, in _get_output_subnet_unit
    line: output, search_choices = cell.get_output(rec_layer=self)
    locals:
      output = <not found>
      search_choices = <not found>
      cell = <local> <TFNetworkRecLayer._SubnetworkRecCell object at 0x7fe37643e7b8>
      cell.get_output = <local> <bound method _SubnetworkRecCell.get_output of <TFNetworkRecLayer._SubnetworkRecCell object at 0x7fe37643e7b8>>
      rec_layer = <not found>
      self = <local> <RecLayer 'output' out_type=Data(shape=(None,), dtype='int32', sparse=True, dim=34908, batch_dim_axis=1, beam_size=12)>
  File "/u/makarov/returnn-hmm-fac/TFNetworkRecLayer.py", line 1459, in get_output
    line: assert fixed_seq_len is not None
    locals:
      fixed_seq_len = <local> None
AssertionError
Unhandled exception <class 'AssertionError'> in thread <_MainThread(MainThread, started 140634717132544)>, proc 22317.

Thread current, main, <_MainThread(MainThread, started 140634717132544)>:
(Excluded thread.)

That were all threads.