Textjuicer - SemANTSum

This is a repository for summarization built as a part of the SemANT project. Textjuicer includes semantsum python package and REST API server both contained in this repository. Further, several language models finetuned for summarization are available from Huggingface. The models are compatible with Transformers library and Ollama.

Installation

This can be installed as a standard python package:

pip install .

To use flash-attention, you need to also install the flash-attn package:

pip install flash-attn

Configuration

The package is built in a way that makes every model fully loadable from a configuration file. The configuration file is a YAML file that contains all the necessary information to load the model.

You can use create_config argument to create a configuration file using the configuration builder:

./run.sh create_config path_to_config.yaml

How to add a new summarization model

For adding an API based summarization model the addition of a new model is as easy as creating a new configuration file and saving it in the config directory.

However, if you need to create a brand-new class for your model, read further.

The module summarizer.py contains abstract base classes for several types of summaries. If there is a new summarization model that does not fit into any of the existing classes, a new class can be added to the module. The new class should inherit from the Summarizer class.

To see an example of how to add a new model, you can inspect the openai.py module.

Also, check the create_config method in the __main__.py and init_summarizer in the summarizator_factory.py module to make your new model compatible with the configuration builder and loader.

Demonstrator

There is a simple command line demonstrator showing how to use API based summarization models.

To run the OpenAI demonstrator, use the run_summarizer argument. You can use one of the configuration files from the config directory.

For these, you also need to configure the API. The API configuration could be provided in the model configuration or using --api_config parameter to specify a separate configuration file for the API.

./run.py run_summarizer semantsum/config/openai_query.yaml --api_config openai_api.yaml

To see an example of a structured (non-plaintext) summary, you can try:

./run.py run_summarizer semantsum/config/openai_query_timeline.yaml --api_config openai_api.yaml 

List available models

There is an argument for listing available models:

./run.py available

An example output:

openai_query (openai)
openai_query_timeline (openai)
openai_query_topic (openai)
openai_single (openai)
ollama_query_tiny_llama (ollama)

Overview of models on Hugging Face

List of models available on Hugging Face:

• https://huggingface.co/BUT-FIT/csmpt-7B-RAGsum
• https://huggingface.co/BUT-FIT/CSTinyLLama-1.2B-RAGsum