Merge remote-tracking branch 'upstream/main' into export_param_fix

Signed-off-by: Boris Fomitchev <[email protected]>
NVIDIA · Sep 1, 2020 · b7e2a47 · b7e2a47
2 parents 7963b4c + 45da106
commit b7e2a47
Show file tree

Hide file tree

Showing 44 changed files with 7,168 additions and 4,606 deletions.
diff --git a/Jenkinsfile b/Jenkinsfile
@@ -247,6 +247,7 @@ pipeline {
             model.train_ds.batch_size=8 \
             model.validation_ds.batch_size=8 \
             trainer.max_epochs=1 \
+            +trainer.max_steps=1 \
             model.language_model.pretrained_model_name=bert-base-uncased \
             model.dataset.version_2_with_negative=false \
             trainer.precision=16 \
@@ -267,6 +268,7 @@ pipeline {
             model.train_ds.batch_size=8 \
             model.validation_ds.batch_size=8 \
             trainer.max_epochs=1 \
+            +trainer.max_steps=1 \
             model.validation_ds.file=/home/TestData/nlp/squad_mini/v2.0/dev-v2.0.json \
             model.language_model.pretrained_model_name=bert-base-uncased \
             model.dataset.version_2_with_negative=true \
@@ -323,6 +325,7 @@ pipeline {
             model.validation_ds.batch_size=4 \
 	    trainer.distributed_backend=ddp \
             trainer.max_epochs=1 \
+            +trainer.max_steps=1 \
             model.validation_ds.file=/home/TestData/nlp/squad_mini/v2.0/dev-v2.0.json \
             model.language_model.pretrained_model_name=megatron-bert-345m-uncased  \
             model.dataset.version_2_with_negative=true \
@@ -356,6 +359,7 @@ pipeline {
             model.train_ds.batch_size=8 \
             model.validation_ds.batch_size=8 \
             trainer.max_epochs=1 \
+            +trainer.max_steps=1 \
             model.validation_ds.file=/home/TestData/nlp/squad_mini/v1.1/dev-v1.1.json \
             model.dataset.do_lower_case=false \
             model.language_model.pretrained_model_name=roberta-base \
@@ -449,7 +453,8 @@ pipeline {
             model.dataset.data_dir=/home/TestData/nlp/glue_fake/MRPC \
             trainer.gpus=[0] \
             +trainer.fast_dev_run=True \
-            exp_manager.exp_dir=examples/nlp/glue_benchmark/mrpc'
+            exp_manager.exp_dir=examples/nlp/glue_benchmark/mrpc \
+            model.output_dir=examples/nlp/glue_benchmark/mrpc'
             sh 'rm -rf examples/nlp/glue_benchmark/mrpc'
           }
         }

diff --git a/README.rst b/README.rst
@@ -98,6 +98,9 @@ To run tutorials:
    * - ASR
      - Online Noise Augmentation
      - `05_Online_Noise_Augmentation.ipynb <https://colab.research.google.com/github/NVIDIA/NeMo/blob/main/tutorials/asr/05_Online_Noise_Augmentation.ipynb>`_
+   * - NLP
+     - Token Classification (Named Entity Recognition)
+     - `Token_Classification_Named_Entity_Recognition_tutorial.ipynb <https://colab.research.google.com/github/NVIDIA/NeMo/blob/main/tutorials/nlp/Token_Classification_Named_Entity_Recognition_tutorial.ipynb>`_
    * - NLP
      - Punctuation and Capitialization
      - `Punctuation_and_Capitalization.ipynb <https://colab.research.google.com/github/NVIDIA/NeMo/blob/main/tutorials/nlp/Punctuation_and_Capitalization.ipynb>`_

diff --git a/docs/source/nlp/models.rst b/docs/source/nlp/models.rst
@@ -1,24 +1,15 @@
 Models
 ======
 
-Currently, NeMo's NLP collection supports the following models:
-
-Supported Tasks and Models:
-
-BERT pretraining
-----------------
-GLUE Benchmark
---------------
-Intent Detection and Slot Filling
----------------------------------
-Text Classification
--------------------
-Name Entity Recognition (NER)
------------------------------
-Punctuation and Capitalization
-------------------------------
-Question Answering
-------------------
-
-Scripts for running these models, can be found under ``<NeMo_Git_root>/example/nlp/``.
-NLP tutorial are located under ``<NeMo_Git_root>/tutorials/nlp/``.
+NeMo's NLP collection supports the following models:
+
+* BERT pretraining
+* GLUE Benchmark
+* Intent Detection and Slot Filling
+* Text Classification
+* Name Entity Recognition (NER)
+* Punctuation and Capitalization
+* Question Answering
+
+Scripts for running these models, could be found under ``NeMo/example/nlp/``.
+NLP tutorials are located under ``NeMo/tutorials/nlp/``.
diff --git a/examples/asr/conf/matchboxnet_3x1x64_v1.yaml b/examples/asr/conf/matchboxnet_3x1x64_v1.yaml
@@ -1,20 +1,24 @@
 name: &name "MatchboxNet-3x1x64-v1"
-sample_rate: &sample_rate 16000
 
 model:
-  timesteps: &timesteps 128
-  repeat: &repeat 1
-  dropout: &dropout 0.0
-  kernel_size_factor: &kfactor 1.0
+  sample_rate: 16000
+  timesteps: 128
+  repeat: 1
+  dropout: 0.0
+  kernel_size_factor: 1.0
 
-  labels: &labels ['bed', 'bird', 'cat', 'dog', 'down', 'eight', 'five', 'four', 'go', 'happy', 'house', 'left', 'marvin',
-                   'nine', 'no', 'off', 'on', 'one', 'right', 'seven', 'sheila', 'six', 'stop', 'three', 'tree', 'two', 'up',
-                   'wow', 'yes', 'zero']
+  labels_full: ['bed', 'bird', 'cat', 'dog', 'down', 'eight', 'five', 'four', 'go', 'happy', 'house', 'left', 'marvin',
+           'nine', 'no', 'off', 'on', 'one', 'right', 'seven', 'sheila', 'six', 'stop', 'three', 'tree', 'two', 'up',
+           'wow', 'yes', 'zero']
+
+  labels_subset: ["yes", "no", "up", "down", "left", "right", "on", "off", "stop", "go", "unknown", "silence"]
+
+  labels: ${model.labels_full}
 
   train_ds:
     manifest_filepath: ???
-    sample_rate: *sample_rate
-    labels: *labels
+    sample_rate: ${model.sample_rate}
+    labels: ${model.labels}
     batch_size: 128
     shuffle: True
     augmentor:
@@ -29,16 +33,16 @@ model:
 
   validation_ds:
     manifest_filepath: ???
-    sample_rate: *sample_rate
-    labels: *labels
+    sample_rate: ${model.sample_rate}
+    labels: ${model.labels}
     batch_size: 128
     shuffle: False
     val_loss_idx: 0
 
   test_ds:
     manifest_filepath: null
-    sample_rate: *sample_rate
-    labels: *labels
+    sample_rate: ${model.sample_rate}
+    labels: ${model.labels}
     batch_size: 128
     shuffle: False
     test_loss_idx: 0
@@ -67,7 +71,7 @@ model:
   crop_or_pad_augment:
     cls: nemo.collections.asr.modules.CropOrPadSpectrogramAugmentation
     params:
-      audio_length: *timesteps
+      audio_length: ${model.timesteps}
 
   encoder:
     cls: nemo.collections.asr.modules.ConvASREncoder
@@ -82,57 +86,57 @@ model:
           kernel: [11]
           stride: [1]
           dilation: [1]
-          dropout: *dropout
+          dropout: ${model.dropout}
           residual: false
           separable: true
-          kernel_size_factor: *kfactor
+          kernel_size_factor: ${model.kernel_size_factor}
 
         - filters: 64
-          repeat: *repeat
+          repeat: ${model.repeat}
           kernel: [13]
           stride: [1]
           dilation: [1]
-          dropout: *dropout
+          dropout: ${model.dropout}
           residual: true
           separable: true
-          kernel_size_factor: *kfactor
+          kernel_size_factor: ${model.kernel_size_factor}
 
         - filters: 64
-          repeat: *repeat
+          repeat: ${model.repeat}
           kernel: [15]
           stride: [1]
           dilation: [1]
-          dropout: *dropout
+          dropout: ${model.dropout}
           residual: true
           separable: true
-          kernel_size_factor: *kfactor
+          kernel_size_factor: ${model.kernel_size_factor}
 
         - filters: 64
-          repeat: *repeat
+          repeat: ${model.repeat}
           kernel: [17]
           stride: [1]
           dilation: [1]
-          dropout: *dropout
+          dropout: ${model.dropout}
           residual: true
           separable: true
-          kernel_size_factor: *kfactor
+          kernel_size_factor: ${model.kernel_size_factor}
 
         - filters: 128
           repeat: 1
           kernel: [29]
           stride: [1]
           dilation: [2]
-          dropout: *dropout
+          dropout: ${model.dropout}
           residual: false
           separable: true
-          kernel_size_factor: *kfactor
+          kernel_size_factor: ${model.kernel_size_factor}
 
         - filters: &enc_final_filters 128
           repeat: 1
           kernel: [1]
           stride: [1]
           dilation: [1]
-          dropout: *dropout
+          dropout: ${model.dropout}
           residual: false
 
   decoder:

diff --git a/examples/asr/conf/matchboxnet_3x1x64_v2.yaml b/examples/asr/conf/matchboxnet_3x1x64_v2.yaml
@@ -1,21 +1,24 @@
 name: &name "MatchboxNet-3x1x64-v2"
 
 model:
-  sample_rate: &sample_rate 16000
+  sample_rate: 16000
+  timesteps: 128
+  repeat: 1
+  dropout: 0.0
+  kernel_size_factor: 1.0
 
-  timesteps: &timesteps 128
-  repeat: &repeat 1
-  dropout: &dropout 0.0
-  kernel_size_factor: &kfactor 1.0
+  labels_full: ['visual', 'wow', 'learn', 'backward', 'dog', 'two', 'left', 'happy', 'nine', 'go', 'up', 'bed', 'stop',
+           'one', 'zero', 'tree', 'seven', 'on', 'four', 'bird', 'right', 'eight', 'no', 'six', 'forward', 'house',
+           'marvin', 'sheila', 'five', 'off', 'three', 'down', 'cat', 'follow', 'yes']
 
-  labels: &labels ['visual', 'wow', 'learn', 'backward', 'dog', 'two', 'left', 'happy', 'nine', 'go', 'up', 'bed', 'stop',
-                   'one', 'zero', 'tree', 'seven', 'on', 'four', 'bird', 'right', 'eight', 'no', 'six', 'forward', 'house',
-                   'marvin', 'sheila', 'five', 'off', 'three', 'down', 'cat', 'follow', 'yes']
+  labels_subset: ["yes", "no", "up", "down", "left", "right", "on", "off", "stop", "go", "unknown", "silence"]
+
+  labels: ${model.labels_full}
 
   train_ds:
     manifest_filepath: ???
-    sample_rate: *sample_rate
-    labels: *labels
+    sample_rate: ${model.sample_rate}
+    labels: ${model.labels}
     batch_size: 128
     shuffle: True
     augmentor:
@@ -30,16 +33,16 @@ model:
 
   validation_ds:
     manifest_filepath: ???
-    sample_rate: *sample_rate
-    labels: *labels
+    sample_rate: ${model.sample_rate}
+    labels: ${model.labels}
     batch_size: 128
     shuffle: False
     val_loss_idx: 0
 
   test_ds:
     manifest_filepath: null
-    sample_rate: *sample_rate
-    labels: *labels
+    sample_rate: ${model.sample_rate}
+    labels: ${model.labels}
     batch_size: 128
     shuffle: False
     test_loss_idx: 0
@@ -68,7 +71,7 @@ model:
   crop_or_pad_augment:
     cls: nemo.collections.asr.modules.CropOrPadSpectrogramAugmentation
     params:
-      audio_length: *timesteps
+      audio_length: ${model.timesteps}
 
   encoder:
     cls: nemo.collections.asr.modules.ConvASREncoder
@@ -83,57 +86,57 @@ model:
           kernel: [11]
           stride: [1]
           dilation: [1]
-          dropout: *dropout
+          dropout: ${model.dropout}
           residual: false
           separable: true
-          kernel_size_factor: *kfactor
+          kernel_size_factor: ${model.kernel_size_factor}
 
         - filters: 64
-          repeat: *repeat
+          repeat: ${model.repeat}
           kernel: [13]
           stride: [1]
           dilation: [1]
-          dropout: *dropout
+          dropout: ${model.dropout}
           residual: true
           separable: true
-          kernel_size_factor: *kfactor
+          kernel_size_factor: ${model.kernel_size_factor}
 
         - filters: 64
-          repeat: *repeat
+          repeat: ${model.repeat}
           kernel: [15]
           stride: [1]
           dilation: [1]
-          dropout: *dropout
+          dropout: ${model.dropout}
           residual: true
           separable: true
-          kernel_size_factor: *kfactor
+          kernel_size_factor: ${model.kernel_size_factor}
 
         - filters: 64
-          repeat: *repeat
+          repeat: ${model.repeat}
           kernel: [17]
           stride: [1]
           dilation: [1]
-          dropout: *dropout
+          dropout: ${model.dropout}
           residual: true
           separable: true
-          kernel_size_factor: *kfactor
+          kernel_size_factor: ${model.kernel_size_factor}
 
         - filters: 128
           repeat: 1
           kernel: [29]
           stride: [1]
           dilation: [2]
-          dropout: *dropout
+          dropout: ${model.dropout}
           residual: false
           separable: true
-          kernel_size_factor: *kfactor
+          kernel_size_factor: ${model.kernel_size_factor}
 
         - filters: &enc_final_filters 128
           repeat: 1
           kernel: [1]
           stride: [1]
           dilation: [1]
-          dropout: *dropout
+          dropout: ${model.dropout}
           residual: false
 
   decoder:

diff --git a/examples/asr/speech_to_label.py b/examples/asr/speech_to_label.py
@@ -56,7 +56,7 @@
     --background_data_root=<path where the background data are stored> \
     --rebalance_method=<'under' or 'over' of 'fixed'> \ 
     --log
-    (Optional --generate (for demonstration in tutorial). If you want to use your own background noise data, make sure to delete --generates)
+    (Optional --demo (for demonstration in tutorial). If you want to use your own background noise data, make sure to delete --demo)
 ```
 
 ## Train to convergence