Skip to content
Permalink
master
Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?

web-data/gluonnlp/logs/bert/finetuned_mrpc.log

Go to file
 
 
Cannot retrieve contributors at this time
305 lines (305 sloc) 15.1 KB
Raw Blame
Open in GitHub Desktop
INFO:root:Namespace(accumulate=None, batch_size=32, dev_batch_size=8, epochs=3, gpu=True, log_interval=10, lr=2e-05, max_len=128, optimizer='bertadam', seed=2, task_name='MRPC', test_batch_size=8, warmup_ratio=0.1)
[11:45:38] src/storage/storage.cc:108: Using GPUPooledRoundedStorageManager.
INFO:root:BERTClassifier(
(bert): BERTModel(
(encoder): BERTEncoder(
(dropout_layer): Dropout(p = 0.1, axes=())
(layer_norm): BERTLayerNorm(eps=1e-12, axis=-1, center=True, scale=True, in_channels=768)
(transformer_cells): HybridSequential(
(0): BERTEncoderCell(
(dropout_layer): Dropout(p = 0.1, axes=())
(attention_cell): MultiHeadAttentionCell(
(_base_cell): DotProductAttentionCell(
(_dropout_layer): Dropout(p = 0.1, axes=())
)
(proj_query): Dense(768 -> 768, linear)
(proj_key): Dense(768 -> 768, linear)
(proj_value): Dense(768 -> 768, linear)
)
(proj): Dense(768 -> 768, linear)
(ffn): BERTPositionwiseFFN(
(ffn_1): Dense(768 -> 3072, linear)
(activation): GELU()
(ffn_2): Dense(3072 -> 768, linear)
(dropout_layer): Dropout(p = 0.1, axes=())
(layer_norm): BERTLayerNorm(eps=1e-12, axis=-1, center=True, scale=True, in_channels=768)
)
(layer_norm): BERTLayerNorm(eps=1e-12, axis=-1, center=True, scale=True, in_channels=768)
)
(1): BERTEncoderCell(
(dropout_layer): Dropout(p = 0.1, axes=())
(attention_cell): MultiHeadAttentionCell(
(_base_cell): DotProductAttentionCell(
(_dropout_layer): Dropout(p = 0.1, axes=())
)
(proj_query): Dense(768 -> 768, linear)
(proj_key): Dense(768 -> 768, linear)
(proj_value): Dense(768 -> 768, linear)
)
(proj): Dense(768 -> 768, linear)
(ffn): BERTPositionwiseFFN(
(ffn_1): Dense(768 -> 3072, linear)
(activation): GELU()
(ffn_2): Dense(3072 -> 768, linear)
(dropout_layer): Dropout(p = 0.1, axes=())
(layer_norm): BERTLayerNorm(eps=1e-12, axis=-1, center=True, scale=True, in_channels=768)
)
(layer_norm): BERTLayerNorm(eps=1e-12, axis=-1, center=True, scale=True, in_channels=768)
)
(2): BERTEncoderCell(
(dropout_layer): Dropout(p = 0.1, axes=())
(attention_cell): MultiHeadAttentionCell(
(_base_cell): DotProductAttentionCell(
(_dropout_layer): Dropout(p = 0.1, axes=())
)
(proj_query): Dense(768 -> 768, linear)
(proj_key): Dense(768 -> 768, linear)
(proj_value): Dense(768 -> 768, linear)
)
(proj): Dense(768 -> 768, linear)
(ffn): BERTPositionwiseFFN(
(ffn_1): Dense(768 -> 3072, linear)
(activation): GELU()
(ffn_2): Dense(3072 -> 768, linear)
(dropout_layer): Dropout(p = 0.1, axes=())
(layer_norm): BERTLayerNorm(eps=1e-12, axis=-1, center=True, scale=True, in_channels=768)
)
(layer_norm): BERTLayerNorm(eps=1e-12, axis=-1, center=True, scale=True, in_channels=768)
)
(3): BERTEncoderCell(
(dropout_layer): Dropout(p = 0.1, axes=())
(attention_cell): MultiHeadAttentionCell(
(_base_cell): DotProductAttentionCell(
(_dropout_layer): Dropout(p = 0.1, axes=())
)
(proj_query): Dense(768 -> 768, linear)
(proj_key): Dense(768 -> 768, linear)
(proj_value): Dense(768 -> 768, linear)
)
(proj): Dense(768 -> 768, linear)
(ffn): BERTPositionwiseFFN(
(ffn_1): Dense(768 -> 3072, linear)
(activation): GELU()
(ffn_2): Dense(3072 -> 768, linear)
(dropout_layer): Dropout(p = 0.1, axes=())
(layer_norm): BERTLayerNorm(eps=1e-12, axis=-1, center=True, scale=True, in_channels=768)
)
(layer_norm): BERTLayerNorm(eps=1e-12, axis=-1, center=True, scale=True, in_channels=768)
)
(4): BERTEncoderCell(
(dropout_layer): Dropout(p = 0.1, axes=())
(attention_cell): MultiHeadAttentionCell(
(_base_cell): DotProductAttentionCell(
(_dropout_layer): Dropout(p = 0.1, axes=())
)
(proj_query): Dense(768 -> 768, linear)
(proj_key): Dense(768 -> 768, linear)
(proj_value): Dense(768 -> 768, linear)
)
(proj): Dense(768 -> 768, linear)
(ffn): BERTPositionwiseFFN(
(ffn_1): Dense(768 -> 3072, linear)
(activation): GELU()
(ffn_2): Dense(3072 -> 768, linear)
(dropout_layer): Dropout(p = 0.1, axes=())
(layer_norm): BERTLayerNorm(eps=1e-12, axis=-1, center=True, scale=True, in_channels=768)
)
(layer_norm): BERTLayerNorm(eps=1e-12, axis=-1, center=True, scale=True, in_channels=768)
)
(5): BERTEncoderCell(
(dropout_layer): Dropout(p = 0.1, axes=())
(attention_cell): MultiHeadAttentionCell(
(_base_cell): DotProductAttentionCell(
(_dropout_layer): Dropout(p = 0.1, axes=())
)
(proj_query): Dense(768 -> 768, linear)
(proj_key): Dense(768 -> 768, linear)
(proj_value): Dense(768 -> 768, linear)
)
(proj): Dense(768 -> 768, linear)
(ffn): BERTPositionwiseFFN(
(ffn_1): Dense(768 -> 3072, linear)
(activation): GELU()
(ffn_2): Dense(3072 -> 768, linear)
(dropout_layer): Dropout(p = 0.1, axes=())
(layer_norm): BERTLayerNorm(eps=1e-12, axis=-1, center=True, scale=True, in_channels=768)
)
(layer_norm): BERTLayerNorm(eps=1e-12, axis=-1, center=True, scale=True, in_channels=768)
)
(6): BERTEncoderCell(
(dropout_layer): Dropout(p = 0.1, axes=())
(attention_cell): MultiHeadAttentionCell(
(_base_cell): DotProductAttentionCell(
(_dropout_layer): Dropout(p = 0.1, axes=())
)
(proj_query): Dense(768 -> 768, linear)
(proj_key): Dense(768 -> 768, linear)
(proj_value): Dense(768 -> 768, linear)
)
(proj): Dense(768 -> 768, linear)
(ffn): BERTPositionwiseFFN(
(ffn_1): Dense(768 -> 3072, linear)
(activation): GELU()
(ffn_2): Dense(3072 -> 768, linear)
(dropout_layer): Dropout(p = 0.1, axes=())
(layer_norm): BERTLayerNorm(eps=1e-12, axis=-1, center=True, scale=True, in_channels=768)
)
(layer_norm): BERTLayerNorm(eps=1e-12, axis=-1, center=True, scale=True, in_channels=768)
)
(7): BERTEncoderCell(
(dropout_layer): Dropout(p = 0.1, axes=())
(attention_cell): MultiHeadAttentionCell(
(_base_cell): DotProductAttentionCell(
(_dropout_layer): Dropout(p = 0.1, axes=())
)
(proj_query): Dense(768 -> 768, linear)
(proj_key): Dense(768 -> 768, linear)
(proj_value): Dense(768 -> 768, linear)
)
(proj): Dense(768 -> 768, linear)
(ffn): BERTPositionwiseFFN(
(ffn_1): Dense(768 -> 3072, linear)
(activation): GELU()
(ffn_2): Dense(3072 -> 768, linear)
(dropout_layer): Dropout(p = 0.1, axes=())
(layer_norm): BERTLayerNorm(eps=1e-12, axis=-1, center=True, scale=True, in_channels=768)
)
(layer_norm): BERTLayerNorm(eps=1e-12, axis=-1, center=True, scale=True, in_channels=768)
)
(8): BERTEncoderCell(
(dropout_layer): Dropout(p = 0.1, axes=())
(attention_cell): MultiHeadAttentionCell(
(_base_cell): DotProductAttentionCell(
(_dropout_layer): Dropout(p = 0.1, axes=())
)
(proj_query): Dense(768 -> 768, linear)
(proj_key): Dense(768 -> 768, linear)
(proj_value): Dense(768 -> 768, linear)
)
(proj): Dense(768 -> 768, linear)
(ffn): BERTPositionwiseFFN(
(ffn_1): Dense(768 -> 3072, linear)
(activation): GELU()
(ffn_2): Dense(3072 -> 768, linear)
(dropout_layer): Dropout(p = 0.1, axes=())
(layer_norm): BERTLayerNorm(eps=1e-12, axis=-1, center=True, scale=True, in_channels=768)
)
(layer_norm): BERTLayerNorm(eps=1e-12, axis=-1, center=True, scale=True, in_channels=768)
)
(9): BERTEncoderCell(
(dropout_layer): Dropout(p = 0.1, axes=())
(attention_cell): MultiHeadAttentionCell(
(_base_cell): DotProductAttentionCell(
(_dropout_layer): Dropout(p = 0.1, axes=())
)
(proj_query): Dense(768 -> 768, linear)
(proj_key): Dense(768 -> 768, linear)
(proj_value): Dense(768 -> 768, linear)
)
(proj): Dense(768 -> 768, linear)
(ffn): BERTPositionwiseFFN(
(ffn_1): Dense(768 -> 3072, linear)
(activation): GELU()
(ffn_2): Dense(3072 -> 768, linear)
(dropout_layer): Dropout(p = 0.1, axes=())
(layer_norm): BERTLayerNorm(eps=1e-12, axis=-1, center=True, scale=True, in_channels=768)
)
(layer_norm): BERTLayerNorm(eps=1e-12, axis=-1, center=True, scale=True, in_channels=768)
)
(10): BERTEncoderCell(
(dropout_layer): Dropout(p = 0.1, axes=())
(attention_cell): MultiHeadAttentionCell(
(_base_cell): DotProductAttentionCell(
(_dropout_layer): Dropout(p = 0.1, axes=())
)
(proj_query): Dense(768 -> 768, linear)
(proj_key): Dense(768 -> 768, linear)
(proj_value): Dense(768 -> 768, linear)
)
(proj): Dense(768 -> 768, linear)
(ffn): BERTPositionwiseFFN(
(ffn_1): Dense(768 -> 3072, linear)
(activation): GELU()
(ffn_2): Dense(3072 -> 768, linear)
(dropout_layer): Dropout(p = 0.1, axes=())
(layer_norm): BERTLayerNorm(eps=1e-12, axis=-1, center=True, scale=True, in_channels=768)
)
(layer_norm): BERTLayerNorm(eps=1e-12, axis=-1, center=True, scale=True, in_channels=768)
)
(11): BERTEncoderCell(
(dropout_layer): Dropout(p = 0.1, axes=())
(attention_cell): MultiHeadAttentionCell(
(_base_cell): DotProductAttentionCell(
(_dropout_layer): Dropout(p = 0.1, axes=())
)
(proj_query): Dense(768 -> 768, linear)
(proj_key): Dense(768 -> 768, linear)
(proj_value): Dense(768 -> 768, linear)
)
(proj): Dense(768 -> 768, linear)
(ffn): BERTPositionwiseFFN(
(ffn_1): Dense(768 -> 3072, linear)
(activation): GELU()
(ffn_2): Dense(3072 -> 768, linear)
(dropout_layer): Dropout(p = 0.1, axes=())
(layer_norm): BERTLayerNorm(eps=1e-12, axis=-1, center=True, scale=True, in_channels=768)
)
(layer_norm): BERTLayerNorm(eps=1e-12, axis=-1, center=True, scale=True, in_channels=768)
)
)
)
(word_embed): HybridSequential(
(0): Embedding(30522 -> 768, float32)
(1): Dropout(p = 0.1, axes=())
)
(token_type_embed): HybridSequential(
(0): Embedding(2 -> 768, float32)
(1): Dropout(p = 0.1, axes=())
)
(pooler): Dense(768 -> 768, Activation(tanh))
)
(classifier): HybridSequential(
(0): Dropout(p = 0.1, axes=())
(1): Dense(None -> 2, linear)
)
)
INFO:root:Setting MXNET_GPU_MEM_POOL_TYPE="Round" may lead to higher memory usage and faster speed. If you encounter OOM errors, please unset this environment variable.
INFO:root:[Epoch 1 Batch 10/119] loss=0.6271, lr=0.0000059, metrics=accuracy:0.6937,f1:0.8171
INFO:root:[Epoch 1 Batch 20/119] loss=0.6428, lr=0.0000118, metrics=accuracy:0.6709,f1:0.7933
INFO:root:[Epoch 1 Batch 30/119] loss=0.5898, lr=0.0000176, metrics=accuracy:0.6828,f1:0.7804
INFO:root:[Epoch 1 Batch 40/119] loss=0.5413, lr=0.0000196, metrics=accuracy:0.7017,f1:0.7784
INFO:root:[Epoch 1 Batch 50/119] loss=0.6055, lr=0.0000190, metrics=accuracy:0.6997,f1:0.7572
INFO:root:[Epoch 1 Batch 60/119] loss=0.5425, lr=0.0000183, metrics=accuracy:0.7070,f1:0.7427
INFO:root:[Epoch 1 Batch 70/119] loss=0.6457, lr=0.0000177, metrics=accuracy:0.7001,f1:0.7210
INFO:root:[Epoch 1 Batch 80/119] loss=0.5171, lr=0.0000170, metrics=accuracy:0.7043,f1:0.7207
INFO:root:[Epoch 1 Batch 90/119] loss=0.4528, lr=0.0000164, metrics=accuracy:0.7117,f1:0.7127
INFO:root:[Epoch 1 Batch 100/119] loss=0.4327, lr=0.0000157, metrics=accuracy:0.7196,f1:0.7091
INFO:root:[Epoch 1 Batch 110/119] loss=0.4692, lr=0.0000151, metrics=accuracy:0.7246,f1:0.7029
INFO:root:validation metrics:accuracy:0.8309,f1:0.7159
INFO:root:Time cost=54.4s
INFO:root:[Epoch 2 Batch 10/119] loss=0.3106, lr=0.0000139, metrics=accuracy:0.8969,f1:0.6467
INFO:root:[Epoch 2 Batch 20/119] loss=0.2445, lr=0.0000132, metrics=accuracy:0.9051,f1:0.6670
INFO:root:[Epoch 2 Batch 30/119] loss=0.2599, lr=0.0000126, metrics=accuracy:0.9023,f1:0.6672
INFO:root:[Epoch 2 Batch 40/119] loss=0.2171, lr=0.0000119, metrics=accuracy:0.9089,f1:0.6631
INFO:root:[Epoch 2 Batch 50/119] loss=0.3426, lr=0.0000113, metrics=accuracy:0.9020,f1:0.6684
INFO:root:[Epoch 2 Batch 60/119] loss=0.3033, lr=0.0000106, metrics=accuracy:0.8976,f1:0.6689
INFO:root:[Epoch 2 Batch 70/119] loss=0.2578, lr=0.0000100, metrics=accuracy:0.8958,f1:0.6712
INFO:root:[Epoch 2 Batch 80/119] loss=0.3022, lr=0.0000093, metrics=accuracy:0.8943,f1:0.6711
INFO:root:[Epoch 2 Batch 90/119] loss=0.2507, lr=0.0000087, metrics=accuracy:0.8924,f1:0.6719
INFO:root:[Epoch 2 Batch 100/119] loss=0.2629, lr=0.0000080, metrics=accuracy:0.8930,f1:0.6765
INFO:root:[Epoch 2 Batch 110/119] loss=0.2550, lr=0.0000074, metrics=accuracy:0.8931,f1:0.6736
INFO:root:validation metrics:accuracy:0.8873,f1:0.7009
INFO:root:Time cost=52.9s
INFO:root:[Epoch 3 Batch 10/119] loss=0.0993, lr=0.0000061, metrics=accuracy:0.9712,f1:0.6634
INFO:root:[Epoch 3 Batch 20/119] loss=0.0745, lr=0.0000055, metrics=accuracy:0.9728,f1:0.6612
INFO:root:[Epoch 3 Batch 30/119] loss=0.1482, lr=0.0000049, metrics=accuracy:0.9682,f1:0.6713
INFO:root:[Epoch 3 Batch 40/119] loss=0.1198, lr=0.0000042, metrics=accuracy:0.9684,f1:0.6792
INFO:root:[Epoch 3 Batch 50/119] loss=0.1264, lr=0.0000036, metrics=accuracy:0.9631,f1:0.6822
INFO:root:[Epoch 3 Batch 60/119] loss=0.0803, lr=0.0000029, metrics=accuracy:0.9652,f1:0.6830
INFO:root:[Epoch 3 Batch 70/119] loss=0.1020, lr=0.0000023, metrics=accuracy:0.9662,f1:0.6776
INFO:root:[Epoch 3 Batch 80/119] loss=0.0679, lr=0.0000016, metrics=accuracy:0.9686,f1:0.6788
INFO:root:[Epoch 3 Batch 90/119] loss=0.1596, lr=0.0000010, metrics=accuracy:0.9661,f1:0.6809
INFO:root:[Epoch 3 Batch 100/119] loss=0.1332, lr=0.0000003, metrics=accuracy:0.9651,f1:0.6849
INFO:root:[Epoch 3 Batch 110/119] loss=0.0906, lr=-0.0000003, metrics=accuracy:0.9655,f1:0.6813
INFO:root:validation metrics:accuracy:0.8873,f1:0.7033
INFO:root:Time cost=52.3s