PICARD - Parsing Incrementally for Constrained Auto-Regressive Decoding from Language Models

Related tags

Deep Learningpicard
Overview


make it parse

build license

This is the official implementation of the following paper:

Torsten Scholak, Nathan Schucher, Dzmitry Bahdanau. PICARD - Parsing Incrementally for Constrained Auto-Regressive Decoding from Language Models. Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (EMNLP).

If you use this code, please cite:

@inproceedings{Scholak2021:PICARD,
  author = {Torsten Scholak and Nathan Schucher and Dzmitry Bahdanau},
  title = {PICARD - Parsing Incrementally for Constrained Auto-Regressive Decoding from Language Models},
  booktitle = {Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing},
  year = {2021},
  publisher = {Association for Computational Linguistics},
}

Overview

This code implements:

  • The PICARD algorithm for constrained decoding from language models.
  • A text-to-SQL semantic parser based on pre-trained sequence-to-sequence models and PICARD achieving state-of-the-art performance on both the Spider and the CoSQL datasets.

About PICARD

TL;DR: We introduce PICARD -- a new method for simple and effective constrained decoding from large pre-trained language models. On the challenging Spider and CoSQL text-to-SQL datasets, PICARD significantly improves the performance of fine-tuned but otherwise unmodified T5 models. Using PICARD, our T5-3B models achieved state-of-the-art performance on both Spider and CoSQL.

In text-to-SQL translation, the goal is to translate a natural language question into a SQL query. There are two main challenges to this task:

  1. The generated SQL needs to be semantically correct, that is, correctly reflect the meaning of the question.
  2. The SQL also needs to be valid, that is, it must not result in an execution error.

So far, there has been a trade-off between these two goals: The second problem can be solved by using a special decoder architecture that -- by construction -- always produces valid SQL. This is the approach taken by most prior work. Those decoders are called "constrained decoders", and they need to be trained from scratch on the text-to-SQL dataset. However, this limits the generality of the decoders, which is a problem for the first goal.

A better approach would be to use a pre-trained encoder-decoder model and to constrain its decoder to produce valid SQL after fine-tuning the model on the text-to-SQL task. This is the approach taken by the PICARD algorithm.

How is PICARD different from existing constrained decoders?

  • It’s an incremental parsing algorithm that integrates with ordinary beam search.
  • It doesn’t require any training.
  • It doesn’t require modifying the model.
  • It works with any model that generates a sequence of tokens (including language models).
  • It doesn’t require a special vocabulary.
  • It works with character-, sub-word-, and word-level language models.

How does PICARD work?

The following picture shows how PICARD is integrated with beam search.



Decoding starts from the left and proceeds to the right. The algorithm begins with a single token (usually <s>), and then keeps expanding the beam with hypotheses generated token-by-token by the decoder. At each decoding step and for each hypothesis, PICARD checks whether the next top-k tokens are valid. In the image above, only 3 token predictions are shown, and k is set to 2. Valid tokens () are added to the beam. Invalid ones (☒) are discarded. The k+1-th, k+2-th, ... tokens are discarded, too. Like in normal beam search, the beam is pruned to contain only the top-n hypotheses. n is the beam size, and in the image above it is set to 2 as well. Hypotheses that are terminated with the end-of-sentence token (usually </s>) are not expanded further. The algorithm stops when the all hypotheses are terminated or when the maximum number of tokens has been reached.

How does PICARD know whether a token is valid?

In PICARD, checking, accepting, and rejecting of tokens and token sequences is achieved through parsing. Parsing means that we attempt to assemble a data structure from the tokens that are currently in the beam or are about to be added to it. This data structure (and the parsing rules that are used to build it) encode the constraints we want to enforce.

In the case of SQL, the data structure we parse to is the abstract syntax tree (AST) of the SQL query. The parsing rules are defined in a computer program called a parser. Database engines, such as PostgreSQL, MySQL, and SQLite, have their own built-in parser that they use internally to process SQL queries. For Spider and CoSQL, we have implemented a parser that supports a subset of the SQLite syntax and that checks additional constraints on the AST. In our implementation, the parsing rules are made up from simpler rules and primitives that are provided by a third-party parsing library.

PICARD uses a parsing library called attoparsec that supports incremental input. This is a special capability that is not available in many other parsing libraries. You can feed attoparsec a string that represents only part of the expected input to parse. When parsing reaches the end of an input fragment, attoparsec will return a continuation function that can be used to continue parsing. Think of the continuation function as a suspended computation that can be resumed later. Input fragments can be parsed one after the other when they become available until the input is complete.

Herein lies the key to PICARD: Incremental parsing of input fragments is exactly what we need to check tokens one by one during decoding.

In PICARD, parsing is initialized with an empty string, and attoparsec will return the first continuation function. We then call that continuation function with all the token predictions we want to check in the first decoding step. For those tokens that are valid, the continuation function will return a new continuation function that we can use to continue parsing in the next decoding step. For those tokens that are invalid, the continuation function will return a failure value which cannot be used to continue parsing. Such failures are discarded and never end up in the beam. We repeat the process until the end of the input is reached. The input is complete once the model predicts the end-of-sentence token. When that happens, we finalize the parsing by calling the continuation function with an empty string. If the parsing is successful, it will return the final AST. If not, it will return a failure value.

The parsing rules are described at a high level in the PICARD paper. For details, see the PICARD code, specifically the Language.SQL.SpiderSQL.Parse module.

How well does PICARD work?

Let's look at the numbers:

On Spider

URL Exact-set Match Accuracy Execution Accuracy
Dev Test Dev Test
tscholak/cxmefzzi w PICARD 75.5 % 71.9 % 79.3 % 75.1 %
tscholak/cxmefzzi w/o PICARD 71.5 % 68.0 % 74.4 % 70.1 %

Click on the links to download the model.

On CoSQL Dialogue State Tracking

URL Question Match Accuracy Interaction Match Accuracy
Dev Test Dev Test
tscholak/2e826ioa w PICARD 56.9 % 54.6 % 24.2 % 23.7 %
tscholak/2e826ioa w/o PICARD 53.8 % 51.4 % 21.8 % 21.7 %

Click on the links to download the model.

Quick Start

Prerequisites

This repository uses git submodules. Clone it like this:

$ git clone [email protected]:ElementAI/picard.git
$ cd picard
$ git submodule update --init --recursive

Training

The training script is located in seq2seq/run_seq2seq.py. You can run it with:

$ make train

The model will be trained on the Spider dataset by default. You can also train on CoSQL by running make train-cosql.

The training script will create the directory train in the current directory. Training artifacts like checkpoints will be stored in this directory.

The default configuration is stored in configs/train.json. The settings are optimized for a GPU with 40GB of memory.

These training settings should result in a model with at least 71% exact-set-match accuracy on the Spider development set. With PICARD, the accuracy should go up to at least 75%.

We have uploaded a model trained on the Spider dataset to the huggingface model hub, tscholak/cxmefzzi. A model trained on the CoSQL dialog state tracking dataset is available, too, tscholak/2e826ioa.

Evaluation

The evaluation script is located in seq2seq/run_seq2seq.py. You can run it with:

$ make eval

By default, the evaluation will be run on the Spider evaluation set. Evaluation on the CoSQL evaluation set can be run with make eval-cosql.

The evaluation script will create the directory eval in the current directory. The evaluation results will be stored there.

The default configuration is stored in configs/eval.json.

Docker

There are three docker images that can be used to run the code:

  • tscholak/text-to-sql-dev: Base image with development dependencies. Use this for development. Pull it with make pull-dev-image from the docker hub. Rebuild the image with make build-dev-image.
  • tsscholak/text-to-sql-train: Training image with development dependencies but without Picard dependencies. Use this for fine-tuning a model. Pull it with make pull-train-image from the docker hub. Rebuild the image with make build-train-image.
  • tscholak/text-to-sql-eval: Training/evaluation image with all dependencies. Use this for evaluating a fine-tuned model with Picard. This image can also be used for training if you want to run evaluation during training with Picard. Pull it with make pull-eval-image from the docker hub. Rebuild the image with make build-eval-image.

All images are tagged with the current commit hash. The images are built with the buildx tool which is available in the latest docker-ce. Use make init-buildkit to initialize the buildx tool on your machine. You can then use make build-dev-image, make build-train-image, etc. to rebuild the images. Local changes to the code will not be reflected in the docker images unless they are committed to git.

Owner
ElementAI
ElementAI
JudeasRx - graphical app for doing personalized causal medicine using the methods invented by Judea Pearl et al.

JudeasRX Instructions Read the references given in the Theory and Notation section below Fire up the Jupyter Notebook judeas-rx.ipynb The notebook dra

Robert R. Tucci 19 Nov 07, 2022
A Comparative Review of Recent Kinect-Based Action Recognition Algorithms (TIP2020, Matlab codes)

A Comparative Review of Recent Kinect-Based Action Recognition Algorithms This repo contains: the HDG implementation (Matlab codes) for 'Analysis and

Lei Wang 5 Oct 22, 2022
Nest Protect integration for Home Assistant. This will allow you to integrate your smoke, heat, co and occupancy status real-time in HA.

Nest Protect integration for Home Assistant Custom component for Home Assistant to interact with Nest Protect devices via an undocumented and unoffici

Mick Vleeshouwer 175 Dec 29, 2022
DeFMO: Deblurring and Shape Recovery of Fast Moving Objects (CVPR 2021)

Evaluation, Training, Demo, and Inference of DeFMO DeFMO: Deblurring and Shape Recovery of Fast Moving Objects (CVPR 2021) Denys Rozumnyi, Martin R. O

Denys Rozumnyi 139 Dec 26, 2022
RealFormer-Pytorch Implementation of RealFormer using pytorch

RealFormer-Pytorch Implementation of RealFormer using pytorch. Includes comparison with classical Transformer on image classification task (ViT) wrt C

Simo Ryu 90 Dec 08, 2022
A Tensorflow implementation of BicycleGAN.

BicycleGAN implementation in Tensorflow As part of the implementation series of Joseph Lim's group at USC, our motivation is to accelerate (or sometim

Cognitive Learning for Vision and Robotics (CLVR) lab @ USC 97 Dec 02, 2022
CVPR 2021 - Official code repository for the paper: On Self-Contact and Human Pose.

TUCH This repo is part of our project: On Self-Contact and Human Pose. [Project Page] [Paper] [MPI Project Page] License Software Copyright License fo

Lea Müller 45 Jan 07, 2023
Portfolio asset allocation strategies: from Markowitz to RNNs

Portfolio asset allocation strategies: from Markowitz to RNNs Research project to explore different approaches for optimal portfolio allocation starti

Luigi Filippo Chiara 1 Feb 05, 2022
A 3D Dense mapping backend library of SLAM based on taichi-Lang designed for the aerial swarm.

TaichiSLAM This project is a 3D Dense mapping backend library of SLAM based Taichi-Lang, designed for the aerial swarm. Intro Taichi is an efficient d

XuHao 230 Dec 19, 2022
Bayesian inference for Permuton-induced Chinese Restaurant Process (NeurIPS2021).

Permuton-induced Chinese Restaurant Process Note: Currently only the Matlab version is available, but a Python version will be available soon! This is

NTT Communication Science Laboratories 3 Dec 17, 2022
Meta-learning for NLP

Self-Supervised Meta-Learning for Few-Shot Natural Language Classification Tasks Code for training the meta-learning models and fine-tuning on downstr

IESL 43 Nov 08, 2022
A python module for configuration of block devices

Blivet is a python module for system storage configuration. CI status Licence See COPYING Installation From Fedora repositories Blivet is available in

78 Dec 14, 2022
Semi-supervised Transfer Learning for Image Rain Removal. In CVPR 2019.

Semi-supervised Transfer Learning for Image Rain Removal This package contains the Python implementation of "Semi-supervised Transfer Learning for Ima

Wei Wei 59 Dec 26, 2022
Repository for open research on optimizers.

Open Optimizers Repository for open research on optimizers. This is a test in sharing research/exploration as it happens. If you use anything from thi

Ariel Ekgren 6 Jun 24, 2022
A GUI for Face Recognition, based upon Docker, Tkinter, GPU and a camera device.

Face Recognition GUI This repository is a GUI version of Face Recognition by Adam Geitgey, where e.g. Docker and Tkinter are utilized. All the materia

Kasper Henriksen 6 Dec 05, 2022
Convolutional Neural Network to detect deforestation in the Amazon Rainforest

Convolutional Neural Network to detect deforestation in the Amazon Rainforest This project is part of my final work as an Aerospace Engineering studen

5 Feb 17, 2022
CoCosNet v2: Full-Resolution Correspondence Learning for Image Translation

CoCosNet v2: Full-Resolution Correspondence Learning for Image Translation (CVPR 2021, oral presentation) CoCosNet v2: Full-Resolution Correspondence

Microsoft 308 Dec 07, 2022
PyTorch for Semantic Segmentation

PyTorch for Semantic Segmentation This repository contains some models for semantic segmentation and the pipeline of training and testing models, impl

Zijun Deng 1.7k Jan 06, 2023
An imperfect information game is a type of game with asymmetric information

DecisionHoldem An imperfect information game is a type of game with asymmetric information. Compared with perfect information game, imperfect informat

Decision AI 25 Dec 23, 2022
Pytorch implementation of BRECQ, ICLR 2021

BRECQ Pytorch implementation of BRECQ, ICLR 2021 @inproceedings{ li&gong2021brecq, title={BRECQ: Pushing the Limit of Post-Training Quantization by Bl

Yuhang Li 148 Dec 28, 2022