Skip to content

HephaestusProject/pytorch-ReCoSa

Repository files navigation

template

Code Coverage Code style: black

Abstract

  • ReCoSA is able to detect relevant contexts and produce a suitable response accordingly. Firstly, a word level LSTM encoder is conducted to obtain the initial representation of each context. Then, the self-attention mechanism is utilized to update both the context and masked response representation. Finally, the attention weights between each context and response representations are computed and used in the further decoding process.

Table (v0.2.4)

  • DSTC7_AVSD
PPL BLEU(4-grams) BLEU(2-grams)
105.34 0.099 0.214

Training history

  • tensorboard 또는 weights & biases를 이용, 학습의 로그의 스크린샷을 올려주세요.

OpenAPI로 Inference 하는 방법

  curl -s "http://127.0.0.1:8000/hello"
  curl -X POST "http://127.0.0.1:8000/model" -H "Content-Type: application/json" -d "{\"input_text\":\"thanks! \"}"

Usage

Environment

  • install from source code
  • dockerfile 이용

Training & Evaluate

./train.sh
./evaluate.sh

Project structure

.
├── Dockerfile
├── LICENSE
├── README.md
├── apply.sh
|── conf
│   ├── api
│   │   ├── ReCoSa.yml
│   │   └── ReCoSa_AVSD.yml
│   ├── dataset
│   │   ├── DSTC7_AVSD.yml
│   │   ├── ubuntu.yml
│   │   └── ubuntu_test.yml
│   ├── model
│   │   ├── ReCoSa.yml
│   │   └── ReCoSa_test.yml
│   ├── tokenizer
│   │   ├── added_tokens.json
│   │   ├── merges.txt
│   │   ├── special_tokens_map.json
│   │   ├── tokenizer_config.json
│   │   └── vocab.json
│   └── trainer
│       ├── ReCoSa.yml
│       └── ReCoSa_test.yml
├── configs
│   └── deploying
│       └── latest.yaml
├── coverage.xml
├── data
│   ├── DSTC7_AVSD
│   │   ├── dial.dev
│   │   ├── dial.test
│   │   └── dial.train
│   ├── DailyDialog
│   │   ├── dial.test
│   │   ├── dial.train
│   │   └── dial.valid
│   ├── PersonaChat
│   │   ├── dial.test
│   │   ├── dial.train
│   │   └── dial.valid
│   └── Ubuntu
│       ├── LICENSE
│       ├── test.csv
│       ├── train.csv
│       └── valid.csv
├── deploying
│   └── helm
│       ├── Chart.yaml
│       ├── templates
│       │   ├── deployment.yaml
│       │   └── service.yaml
│       └── values.yaml
├── evaluate.py
├── evaluate.sh
├── infer.py
├── lightning_logs
├── requirements.txt
├── server.Dockerfile
├── serving
│   ├── __init__.py
│   └── app_factory.py
├── serving
├── LICENSE
├── README.md
├── requirements.txt
├── src
│   ├── core
│   │   └── build_data.py
│   ├── data.py
│   ├── metric.py
│   ├── model
│   │   ├── net.py
│   │   └── ops.py
│   ├── utils
│   │   └── prepare.py
│   └── utils.py
├── tests
│   ├── resources
│   │   └── Ubuntu
│   │       └── sample.csv
│   ├── __init__.py
│   ├── test_api.py
│   ├── test_data.py
│   ├── test_download.py
│   ├── test_inference.py
│   ├── test_model.py
│   └── test_trainer.py
├── train.py
└── train.sh

License

  • Licensed under an MIT license.

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Contributors 3

  •  
  •  
  •