ConvBERT-Prod

Overview

ConvBERT

目录

0. 仓库结构

root:[./]
|--convbert_base_outputs
|      |--args.json
|      |--best.pdparams
|      |      |--model_config.json
|      |      |--model_state.pdparams
|      |      |--tokenizer_config.json
|      |      |--vocab.txt
|--convbert_infer
|      |--inference.pdiparams
|      |--inference.pdiparams.info
|      |--inference.pdmodel
|      |--tokenizer_config.json
|      |--vocab.txt
|--deploy
|      |--inference_python
|      |      |--infer.py
|      |      |--README.md
|      |--serving_python
|      |      |--config.yml
|      |      |--convbert_client
|      |      |      |--serving_client_conf.prototxt
|      |      |      |--serving_client_conf.stream.prototxt
|      |      |--convbert_server
|      |      |      |--inference.pdiparams
|      |      |      |--inference.pdmodel
|      |      |      |--serving_server_conf.prototxt
|      |      |      |--serving_server_conf.stream.prototxt
|      |      |--PipelineServingLogs
|      |      |      |--pipeline.log
|      |      |      |--pipeline.log.wf
|      |      |      |--pipeline.tracer
|      |      |--pipeline_http_client.py
|      |      |--ProcessInfo.json
|      |      |--README.md
|      |      |--web_service.py
|--images
|      |--convbert_framework.jpg
|      |--py_serving_client_results.jpg
|      |--py_serving_startup_visualization.jpg
|--LICENSE
|--output_inference_engine.npy
|--output_predict_engine.npy
|--paddlenlp
|--print_project_tree.py
|--README.md
|--requirements.txt
|--shell
|      |--export.sh
|      |--inference_python.sh
|      |--predict.sh
|      |--train.sh
|      |--train_dist.sh
|--test_tipc
|      |--common_func.sh
|      |--configs
|      |      |--ConvBERT
|      |      |      |--model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
|      |      |      |--train_infer_python.txt
|      |--docs
|      |      |--test_serving.md
|      |      |--test_train_inference_python.md
|      |      |--tipc_guide.png
|      |      |--tipc_serving.png
|      |      |--tipc_train_inference.png
|      |--output
|      |      |--python_infer_cpu_usemkldnn_False_threads_null_precision_null_batchsize_null.log
|      |      |--python_infer_gpu_usetrt_null_precision_null_batchsize_null.log
|      |      |--results_python.log
|      |      |--results_serving.log
|      |      |--server_infer_gpu_pipeline_http_usetrt_null_precision_null_batchsize_1.log
|      |--README.md
|      |--test_serving.sh
|      |--test_train_inference_python.sh
|--tools
|      |--export_model.py
|      |--predict.py
|--train.log
|--train.py

1. 简介

论文: ConvBERT: Improving BERT with Span-based Dynamic Convolution

摘要: 像BERT及其变体这样的预训练语言模型最近在各种自然语言理解任务中取得了令人印象深刻的表现。然而,BERT严重依赖全局自注意力块,因此需要大量内存占用和计算成本。 虽然它的所有注意力头从全局角度查询整个输入序列以生成注意力图,但我们观察到一些头只需要学习局部依赖,这意味着存在计算冗余。 因此,我们提出了一种新颖的基于跨度的动态卷积来代替这些自注意力头,以直接对局部依赖性进行建模。新的卷积头与其余的自注意力头一起形成了一个新的混合注意力块,在全局和局部上下文学习中都更有效。 我们为 BERT 配备了这种混合注意力设计并构建了一个ConvBERT模型。实验表明,ConvBERT 在各种下游任务中明显优于BERT及其变体,具有更低的训练成本和更少的模型参数。 值得注意的是,ConvBERT-base 模型达到86.4GLUE分数,比ELECTRA-base高0.7,同时使用不到1/4的训练成本。

2. 数据集和复现精度

数据集为SST-2

模型 sst-2 dev acc (复现精度)
ConvBERT 0.9461

3. 准备环境与数据

3.1 准备环境

  • 下载代码
git clone https://github.com/junnyu/ConvBERT-Prod.git
  • 安装paddlepaddle
# 需要安装2.2及以上版本的Paddle,如果
# 安装GPU版本的Paddle
pip install paddlepaddle-gpu==2.2.0
# 安装CPU版本的Paddle
pip install paddlepaddle==2.2.0

更多安装方法可以参考:Paddle安装指南

  • 安装requirements
pip install -r requirements.txt

3.2 准备数据

SST-2数据已经集成在paddlenlp仓库中。

3.3 准备模型

如果您希望直接体验评估或者预测推理过程,可以直接根据第2章的内容下载提供的预训练模型,直接体验模型评估、预测、推理部署等内容。

4. 开始使用

4.1 模型训练

  • 单机单卡训练
export CUDA_VISIBLE_DEVICES=0
python -m paddle.distributed.launch --gpus "0" train.py \
    --model_type convbert \
    --model_name_or_path convbert-base \
    --task_name sst-2 \
    --max_seq_length 128 \
    --learning_rate 1e-4 \
    --num_train_epochs 3 \
    --output_dir ./convbert_base_outputs/ \
    --logging_steps 100 \
    --save_steps 400 \
    --batch_size 32   \
    --warmup_proportion 0.1

部分训练日志如下所示。

====================================================================================================
global step 2500/6315, epoch: 1, batch: 394, rank_id: 0, loss: 0.140546, lr: 0.0000671182, speed: 3.7691 step/s
global step 2600/6315, epoch: 1, batch: 494, rank_id: 0, loss: 0.062813, lr: 0.0000653589, speed: 4.1413 step/s
global step 2700/6315, epoch: 1, batch: 594, rank_id: 0, loss: 0.051268, lr: 0.0000635996, speed: 4.1867 step/s
global step 2800/6315, epoch: 1, batch: 694, rank_id: 0, loss: 0.133289, lr: 0.0000618403, speed: 4.1769 step/s
eval loss: 0.342346, acc: 0.9461009174311926,
eval done total : 1.9056718349456787 s
====================================================================================================
  • 单机多卡训练
export CUDA_VISIBLE_DEVICES=0,1,2,3
python -m paddle.distributed.launch --gpus "0,1,2,3" train.py \
    --model_type convbert \
    --model_name_or_path convbert-base \
    --task_name sst-2 \
    --max_seq_length 128 \
    --learning_rate 1e-4 \
    --num_train_epochs 3 \
    --output_dir ./convbert_base_outputs/ \
    --logging_steps 100 \
    --save_steps 400 \
    --batch_size 32   \
    --warmup_proportion 0.1

更多配置参数可以参考train.pyget_args_parser函数。

4.2 模型评估

该项目中,训练与评估脚本同时进行,请查看训练过程中的评价指标。

4.3 模型预测

  • 使用GPU预测
python tools/predict.py --model_path=./convbert_base_outputs/best.pdparams

对于下面的文本进行预测

the problem , it is with most of these things , is the script .

最终输出结果为label_id: 0, prob: 0.9959235191345215,表示预测的标签ID是0,置信度为0.9959

  • 使用CPU预测
python tools/predict.py --model_path=./convbert_base_outputs/best.pdparams --device=cpu

对于下面的文本进行预测

the problem , it is with most of these things , is the script .

最终输出结果为label_id: 0, prob: 0.995919406414032,表示预测的标签ID是0,置信度为0.9959

5. 模型推理部署

5.1 基于Inference的推理

Inference推理教程可参考:链接

5.2 基于Serving的服务化部署

Serving部署教程可参考:链接

6. TIPC自动化测试脚本

以Linux基础训练推理测试为例,测试流程如下。

  • 运行测试命令
bash test_tipc/test_train_inference_python.sh test_tipc/configs/ConvBERT/train_infer_python.txt whole_train_whole_infer

如果运行成功,在终端中会显示下面的内容,具体的日志也会输出到test_tipc/output/文件夹中的文件中。

�[33m Run successfully with command - python train.py --save_steps 400      --max_steps=6315           !  �[0m
�[33m Run successfully with command - python tools/export_model.py --model_path=./convbert_base_outputs/best.pdparams --save_inference_dir ./convbert_infer      !  �[0m
�[33m Run successfully with command - python deploy/inference_python/infer.py --model_dir ./convbert_infer --use_gpu=True               > ./test_tipc/output/python_infer_gpu_usetrt_null_precision_null_batchsize_null.log 2>&1 !  �[0m
�[33m Run successfully with command - python deploy/inference_python/infer.py --model_dir ./convbert_infer --use_gpu=False --benchmark=False               > ./test_tipc/output/python_infer_cpu_usemkldnn_False_threads_null_precision_null_batchsize_null.log 2>&1 !  �[0m

7. 注意

为了可以使用静态图导出功能,本项目修改了paddlenlp仓库中的convbert模型,主要修改部分如下。

    1. 使用paddle.shape而不是tensor.shape获取tensor的形状。
    1. F.unfold对于静态图不怎么友好,只好采用for循环。
if self.conv_type == "sdconv":
    bs = paddle.shape(q)[0]
    seqlen = paddle.shape(q)[1]
    mixed_key_conv_attn_layer = self.key_conv_attn_layer(query)
    conv_attn_layer = mixed_key_conv_attn_layer * q

    # conv_kernel_layer
    conv_kernel_layer = self.conv_kernel_layer(conv_attn_layer)
    conv_kernel_layer = tensor.reshape(
        conv_kernel_layer, shape=[-1, self.conv_kernel_size, 1])
    conv_kernel_layer = F.softmax(conv_kernel_layer, axis=1)
    conv_out_layer = self.conv_out_layer(query)
    conv_out_layer = paddle.stack(
        [
            paddle.slice(F.pad(conv_out_layer, pad=[
                            self.padding, self.padding], data_format="NLC"), [1], starts=[i], ends=[i+seqlen])
            for i in range(self.conv_kernel_size)
        ],
        axis=-1,
    )
    conv_out_layer = tensor.reshape(
        conv_out_layer,
        shape=[-1, self.head_dim, self.conv_kernel_size])
    conv_out_layer = tensor.matmul(conv_out_layer, conv_kernel_layer)
    conv_out = tensor.reshape(
        conv_out_layer,
        shape=[bs, seqlen, self.num_heads, self.head_dim])

8. LICENSE

本项目的发布受Apache 2.0 license许可认证。

9. 参考链接与文献

TODO

Owner
yujun
Please show me your code.
yujun
Meta learning algorithms to train cross-lingual NLI (multi-task) models

Meta learning algorithms to train cross-lingual NLI (multi-task) models

M.Hassan Mojab 4 Nov 20, 2022
Dual languaged (rus+eng) tool for packing and unpacking archives of Silky Engine.

SilkyArcTool English Dual languaged (rus+eng) GUI tool for packing and unpacking archives of Silky Engine. It is not the same arc as used in Ai6WIN. I

Tester 5 Sep 15, 2022
NLPIR tutorial: pretrain for IR. pre-train on raw textual corpus, fine-tune on MS MARCO Document Ranking

pretrain4ir_tutorial NLPIR tutorial: pretrain for IR. pre-train on raw textual corpus, fine-tune on MS MARCO Document Ranking 用作NLPIR实验室, Pre-training

ZYMa 12 Apr 07, 2022
ADCS - Automatic Defect Classification System (ADCS) for SSMC

Table of Contents Table of Contents ADCS Overview Summary Operator's Guide Demo System Design System Logic Training Mode Production System Flow Folder

Tam Zher Min 2 Jun 24, 2022
MEDIALpy: MEDIcal Abbreviations Lookup in Python

A small python package that allows the user to look up common medical abbreviations.

Aberystwyth Systems Biology 7 Nov 09, 2022
official ( API ) for the zAmericanEnglish app in [ Google play ] and [ App store ]

official ( API ) for the zAmericanEnglish app in [ Google play ] and [ App store ]

Plugin 3 Jan 12, 2022
The code for the Subformer, from the EMNLP 2021 Findings paper: "Subformer: Exploring Weight Sharing for Parameter Efficiency in Generative Transformers", by Machel Reid, Edison Marrese-Taylor, and Yutaka Matsuo

Subformer This repository contains the code for the Subformer. To help overcome this we propose the Subformer, allowing us to retain performance while

Machel Reid 10 Dec 27, 2022
An assignment from my grad-level data mining course demonstrating some experience with NLP/neural networks/Pytorch

NLP-Pytorch-Assignment An assignment from my grad-level data mining course (before I started personal projects) demonstrating some experience with NLP

David Thorne 0 Feb 06, 2022
[EMNLP 2021] LM-Critic: Language Models for Unsupervised Grammatical Error Correction

LM-Critic: Language Models for Unsupervised Grammatical Error Correction This repo provides the source code & data of our paper: LM-Critic: Language M

Michihiro Yasunaga 98 Nov 24, 2022
Contract Understanding Atticus Dataset

Contract Understanding Atticus Dataset This repository contains code for the Contract Understanding Atticus Dataset (CUAD), a dataset for legal contra

The Atticus Project 273 Dec 17, 2022
A repo for open resources & information for people to succeed in PhD in CS & career in AI / NLP

A repo for open resources & information for people to succeed in PhD in CS & career in AI / NLP

420 Dec 28, 2022
Transformers implementation for Fall 2021 Clinic

Installation Download miniconda3 if not already installed You can check by running typing conda in command prompt. Use conda to create an environment

Aakash Tripathi 1 Oct 28, 2021
Scikit-learn style model finetuning for NLP

Scikit-learn style model finetuning for NLP Finetune is a library that allows users to leverage state-of-the-art pretrained NLP models for a wide vari

indico 665 Dec 17, 2022
Code for ACL 2021 main conference paper "Conversations are not Flat: Modeling the Intrinsic Information Flow between Dialogue Utterances".

Conversations are not Flat: Modeling the Intrinsic Information Flow between Dialogue Utterances This repository contains the code and pre-trained mode

ICTNLP 90 Dec 27, 2022
Code to reproduce the results of the paper 'Towards Realistic Few-Shot Relation Extraction' (EMNLP 2021)

Realistic Few-Shot Relation Extraction This repository contains code to reproduce the results in the paper "Towards Realistic Few-Shot Relation Extrac

Bloomberg 8 Nov 09, 2022
Constituency Tree Labeling Tool

Constituency Tree Labeling Tool The purpose of this package is to solve the constituency tree labeling problem. Look from the dataset labeled by NLTK,

张宇 6 Dec 20, 2022
Uncomplete archive of files from the European Nopsled Team

European Nopsled CTF Archive This is an archive of collected material from various Capture the Flag competitions that the European Nopsled team played

European Nopsled 4 Nov 24, 2021
A very simple framework for state-of-the-art Natural Language Processing (NLP)

A very simple framework for state-of-the-art NLP. Developed by Humboldt University of Berlin and friends. IMPORTANT: (30.08.2020) We moved our models

flair 12.3k Dec 31, 2022
A simple version of DeTR

DeTR-Lite A simple version of DeTR Before you enjoy this DeTR-Lite The purpose of this project is to allow you to learn the basic knowledge of DeTR. P

Jianhua Yang 11 Jun 13, 2022
IEEEXtreme15.0 Questions And Answers

IEEEXtreme15.0 Questions And Answers IEEEXtreme is a global challenge in which teams of IEEE Student members – advised and proctored by an IEEE member

Dilan Perera 15 Oct 24, 2022