Towards AI Can Help your Team Adopt AI: Corporate Training, Consulting, and Talent Solutions.


The NLP Cypher | 03.21.21
Latest   Machine Learning   Newsletter

The NLP Cypher | 03.21.21

Last Updated on July 24, 2023 by Editorial Team

Author(s): Ricky Costa

Originally published on Towards AI.



The NLP Cypher U+007C 03.21.21

The Field of Reeds

Hey, welcome back! Let’s kick off the newsletter with a lucid sonic dream from a very crafty GAN U+1F635U+1F344. But wait, what is that exactly? For direct experience check out the video below. TL;DR: a savvy dev trained a GAN to generate acid-trip art that transitions in sync with music. This isn’t NLP related but hey, still cool! Here’s the library, it runs on TF-v1, now go get a wav file and play some Pink Floyd:



Lucid Sonic Dreams syncs GAN-generated visuals to music. By default, it uses NVLabs StyleGAN2, with pre-trained models…

Surprise Colab:

Google Colaboratory

Edit description

If you enjoy this read, please give it a U+1F44FU+1F44F and share with friends… U+1F60E

TensorFlow Transformers Library U+1F440

pip install tf-transformers

Just when you thought you’ve seen it all, a library comes along that augments the capability of TensorFlow v2 focusing on NLP. The authors make impressive claims, here are the highlights (from their repo):

  • Faster Auto Regressive Decoding using Tensorflow2. Faster than PyTorch in most experiments (V100 GPU). 80% faster compared to existing TF based libraries (relative difference) Refer benchmark code.
  • Complete TFlite support for BERT, RoBERTA, T5, Albert, mt5 for all down stream tasks except text-generation.
  • Faster sentence-piece alignment (no more LCS overhead).
  • Variable batch text generation for Encoder only models like GPT2.
  • No more hassle of writing long codes for TFRecords. minimal and simple.
  • Off the shelf support for auto-batching or tf.ragged tensors.
  • Pass dictionary outputs directly to loss functions inside using model.compile2 . Refer examples or blog.
  • Multiple mask modes like causal, user-defined, prefix by changing one argument . Refer examples or blog.

Model Support: ALBERT, BERT, RoBERTa, GPT-2, MT5, ELECTRA and T5.

They even include code to switch from Hugging Face to their library: MORTAL KOMBAT U+1F976…


State of the art faster Natural Language Processing in Tensorflow 2.0 . – legacyai/tf-transformers



tf-transformers: faster and easier state-of-the-art NLP in TensorFlow 2.0 tf-transformers is designed to harness the…

C4: 800GB of English Text Released Into the Wild

Google’s C4 dataset has been in a full quarantine. They never offered the option to download it, we could only replicate it. However AllenNLP came along to save the day, and they brought receipts.

They have 3 variants:

  • en: 800GB in TFDS format, 300GB in JSON format
  • en.noclean: 6.3TB in TFDS format, 2.3TB in JSON format
  • realnewslike: 38GB in TFDS format, 15GB in JSON format

Download the C4 dataset! · Discussion #5056 · allenai/allennlp

You can't perform that action at this time. You signed in with another tab or window. You signed out in another tab or…

Multi-Lingual CLIP

How about a multi-lingual CLIP?

How about 101 source languages using multi-lingual BERT (distilled or base)?

This library offers this, and includes an intuitive Colab to test inference. U+1F525U+1F525

Feel free to contact the authors if you want add a language that is currently not supported. This is great work.


Colab Notebook · Pre-trained Models · Report Bug OpenAI recently released the paper Learning Transferable Visual Models…

Almond V.2 Stanford’s Open Sourced Voice Assistant

Want to help develop an open-sourced voice assistant? Almond is here:


  • Spotify (music)
  • Home Assistant (IoT)
  • Weather
  • Jokes
  • Local restaurants
  • FAQs about the assistant itself

Call for testing: Almond 2.0 Alpha

Hello all! I am very pleased to announce the immediate availability of the new alpha release of Almond. This is the…

A100s Bonanza

Google Cloud Platform update, if you want to go nuts with distributed computing you can now get 16 A100s on a single A2 instance. This is serious horsepower, if you want to know what the upper bound of compute looks like on GCP, this is it.

A2 VMs with NVIDIA A100 GPUs are GA U+007C Google Cloud Blog

A2 VMs with NVIDIA A100 GPUs are now generally available for your most demanding workloads including machine learning…

Getting Started with Rust

It’s like Python but fast U+1F601. If you want to know the current state of Rust, check out the Stack Overflow blog U+1F447 . It highlights how the small but devoted Rust community continues to show strength and includes links to tutorials.

Getting started with … Rust – Stack Overflow Blog

In this series, we look at the most loved languages according to the Stack Overflow developer survey, the spread and…

TensorFlow and Quantum: 1 Year Through the Looking Glass

An overview of the Quantum TF library found here:


TensorFlow Quantum (TFQ) is a Python framework for hybrid quantum-classical machine learning that is primarily focused…


TensorFlow Quantum turns one year old

March 18, 2021 — Posted by Michael Broughton, Alan Ho, Masoud Mohseni Last year we announced TensorFlow Quantum (TFQ)…

Software Updates U+1F4BB

Sentence Transformers Update


Transformers 4.4.2

Releases · huggingface/transformers

Two new models are released as part of the S2T implementation: Speech2TextModel and…

AresDB — GPU Powered Real-Time Storage and Query Engine (in GoLang)

From Uber’s engineering group, this is the source code for AresDB. You get to hook up GPUs for super low latency database querying.


AresDB is a GPU-powered real-time analytics storage and query engine. It features low query latency, high data…

Original release blog:

Introducing AresDB: Uber's GPU-Powered Open Source, Real-time Analytics Engine

At Uber, real-time analytics allow us to attain business insights and operational efficiency, enabling us to make…


An awesome new speech library with several pretrained models on the HF repo. Tasks it supports: speaker recognition, speech identification and speech diarization. It runs on PyTorch.

SpeechBrain: A PyTorch Speech Toolkit

SpeechBrain is an open-source and all-in-one speech toolkit. It is designed to be simple, extremely flexible, and…

MongoDB for Storing and Retrieving ML Models

Tutorial w/ code:

How to Use MongoDB to Store and Retrieve ML Models — Python Simplified

If you are looking for a database for storing your machine learning models then this article is for you. You … How to…

The Mega (Incredible) PyTorch Repo

“This is a curated list of tutorials, projects, libraries, videos, papers, books and anything related to the incredible PyTorch.”

This is a HUGE index.


This is a curated list of tutorials, projects, libraries, videos, papers, books and anything related to the incredible…

Repo Cypher U+1F468‍U+1F4BB

A collection of recently released repos that caught our U+1F441

CARTON Transformer

aka: Context Transformer with Stacked Pointer Networks for Conversational Question Answering over Knowledge Graphs…

Performs multi-task semantic parsing for handling the problem of conversational question answering over a large-scale knowledge graph.


Neural semantic parsing approaches have been widely used for Question Answering (QA) systems over knowledge graphs…

Connected Papers U+1F4C8

Multilingual Code-Switching for Zero-Shot Cross-Lingual Intent Prediction and Slot Filling

Train a joint intent prediction and slot filling model in English and generalize to other languages.


Goal: Train a joint intent prediction and slot fillinf model using English and generalize to other languages…

Connected Papers U+1F4C8

BERTRL: Inductive Relation Prediction by BERT

Using BERT to do knowledge base completion.


Code and Data for Paper [Inductive Relation Prediction by BERT], this paper proposes an algorithm to do knowledge base…

Connected Papers U+1F4C8

Multimodal End-to-End Sparse Model for Emotion Recognition


Paper accepted at the NAACL 2021: Multimodal End-to-End Sparse Model for Emotion Recognition, by Wenliang Dai * …

Connected Papers U+1F4C8


Using Transformers to fine-tune the VitaminC dataset for fact verification.


This repository contains the dataset and model for the NAACL 2021 paper: Get Your Vitamin C! Robust Fact Verification…

Connected Papers U+1F4C8


ENtity-CONstrained insertion TransformER, a language model to help improve fine control of content generation i.e. a method for dealing with entity constraints.


Implementation of 2021 EACL paper Enconter. Contribute to LARC-CMU-SMU/Enconter development by creating an account on…

Connected Papers U+1F4C8

Dataset of the Week: ParaQA

What is it?

A question answering dataset with paraphrase responses for single-turn conversation.

Dataset contains 5,000 question-answer pairs with a minimum of two and a maximum of eight unique paraphrased responses for each question.


Where is it?


The dataset was created using a semi-automated framework for generating diverse paraphrasing of the answers using…

Every Sunday we do a weekly round-up of NLP news and code drops from researchers around the world.

For complete coverage, follow our Twitter: @Quantum_Stat

Quantum Stat

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI

Feedback ↓