NGC | Catalog
CatalogCollectionsTLT - Conversational AI

TLT - Conversational AI

Logo for TLT - Conversational AI
Transfer Learning Toolkit (TLT) is a python based AI toolkit for taking purpose-built pre-trained AI models and customizing them with your own data
April 4, 2023
Sorry, your browser does not support inline SVG.
Helm Charts
Sorry, your browser does not support inline SVG.
Sorry, your browser does not support inline SVG.
Sorry, your browser does not support inline SVG.

What is Transfer Learning Toolkit?

Transfer Learning Toolkit (TLT) is a python based AI toolkit for taking purpose-built pre-trained AI models and customizing them with your own data. TLT adapts popular network architectures and backbones to your data, allowing you to train, fine-tune, and export highly optimized and accurate AI models for deployment.

The pre-trained models accelerate the AI training process and reduce costs associated with large scale data collection, labeling, and training models from scratch.

Build end-to-end services and solutions for conversational AI using TLT and Riva. TLT can train models for common conversational AI tasks such as text classification, question answering, speech recognition, and others.

Purpose-built Pre-Trained Models

Purpose-built pre-trained models offer highly accurate AI for a variety of conversational AI tasks. Developers, system builders, and software partners building conversational AI applications can bring their own custom data to train and fine-tune with using these models, instead of going through the hassle of building a large data collection and training from scratch.

The purpose-built models are available on NGC. Under each model card, there is a version that can be deployed as is and a version which can be used with TLT to fine-tune with your own dataset. Models here can be used for Automatic Speech Recognition (ASR), Question Answering (QA), Domain Classification, Named Entity Recognition (NER), Punctuation and Capitalization, and Joint Intent and Slot Classification. See each model card for information about how to fine-tune and evaluate on your data.

Model Accuracy Metric(s) Use Case
ASR: Jasper (English) 3.74%/10.21% WER (LibriSpeech dev-clean/dev-other) English speech recognition
ASR: QuartzNet (English) 4.38%/11.30% WER (LibriSpeech dev-clean/dev-other) English speech recognition (smaller model)
ASR: Citrinet (English) 4.38%/11.30% WER (LibriSpeech dev-clean/dev-other) English speech recognition
QA: Bert Base (SQuAD2.0) 73.35% EM score, 76.44 F1 score Question answering
QA: Bert Large (SQuAD2.0) 77.16% EM score, 80.22% F1 score Question answering
QA: Bert Megatron (SQuAD2.0) 78.0% EM score, 81.35% F1 score Question answering
Domain Classification: BERT 90% accuracy for 4 domains of the weather chatbot Text classification problems (e.g. sentiment analysis, domain detection)
Punctuation and Capitalization: BERT 77% F1 score Punctuation and capitalization of ASR output
NER: BERT 74.21% F1 score Named entity recognition and other token-level classification tasks
Joint Intent and Slot Classification: BERT 95% intent accuracy, 93% slot accuracy Classifying intent and detecting relevant slots in a query

Running Transfer Learning Toolkit

  1. Setup your python environment using python virtualenv and virtualenvwrapper.

  2. In TLT3.0, we have created an abstraction above the container, you will launch all your training jobs from the launcher. No need to manually pull the appropriate container, tlt-launcher will handle that. You may install the launcher using pip with the following commands.

pip3 install nvidia-pyindex
pip3 install nvidia-tlt
  1. Download one of the Jupyter notebooks that you are interested in from NGC resources. For each task, there is a training notebook as well as a deployment notebook. After installing the pre-requisites, all the training/deployment steps will be run from inside the Jupyter notebook.
Conversational AI Task Jupyter Notebooks
Automatic Speech Recognition Resources
Automatic Speech Recognition - Citrinet Resources
Question Answering Resources
Text Classification Resources
Named Entity Recognition Resources
Punctuation and Capitalization Resources
Intent and Slot Classification Resources

Using TLT Pre-trained Models

Pre-trained models for each Conversational AI task can be found under their respective collections here:


By pulling and using the Transfer Learning Tookit for Conversational AI container, you accept the terms and conditions of this license. By downloading and using the models and resources packaged with TLT Conversational AI, you would be accepting the terms of the Riva license.

Technical blogs

Suggested reading

Ethical AI

NVIDIA’s platforms and application frameworks enable developers to build a wide array of AI applications. Consider potential algorithmic bias when choosing or creating the models being deployed. Work with the model’s developer to ensure that it meets the requirements for the relevant industry and use case; that the necessary instruction and documentation are provided to understand error rates, confidence intervals, and results; and that the model is being used under the conditions and in the manner intended.