Skip to content

Latest commit

 

History

History
68 lines (42 loc) · 1.56 KB

README.md

File metadata and controls

68 lines (42 loc) · 1.56 KB
Raw

SyncTOD: Synergizing In-context Learning with Hints for End-to-end Task-oriented Dialog Systems

This is an official repository for the paper - Synergizing In-context Learning with Hints for End-to-end Task-oriented Dialog Systems.

Code Structure

src/ contains source code to train SyncTOD.

data/ contains dataset used to evaluate SyncTOD.

Installation

  1. Clone the repository:

    git clone https://github.com/dair-iitd/SyncTOD.git

  2. Navigate to the source directory:

    cd SyncTOD/src

  3. Install the required dependencies:

    pip install -r requirements.txt

Usage

Once the dependencies are installed, you can start training or evaluating the system. All our code is run from the src directory.

Use following commands to train SyncTOD and generate the prompts.

  1. MultiWOZ:

    bash run_multiwoz.sh

  2. SMD:

    bash run_smd.sh

  3. BiTOD:

    bash run_bitod.sh

Above command stores the test prompts in prompts.json file in respective dataset folders.

We use Azure OpenAI batch API to run the prompts.

python -u submit_job.py --prompt_file=../data/<datast>/prompts.json --model=<azure_endpoint> --max_output_len=256

We evaluate the results using following command.

python -m commons.metrics --dataset=<datast> --pred_path=<batch_api_result.jsonl> --data_path=../data/<dataset>/test.json --entity_file=../data/<dataset>/entities.json