Practical overview
When it fits
Use UniRE when you need research code for joint entity-and-relation extraction and can work with Python scripts, prepared datasets and model checkpoints.
Who it is for
- NLP researchers evaluating joint entity and relation extraction.
- Developers preparing or fine-tuning extraction models on ACE2004, ACE2005 or SciERC.
- Teams seeking an ACE2005 pretrained model and a documented inference example.
Documented uses
- Prepare ACE2004, ACE2005 or SciERC data for the repository's model workflow.
- Fine-tune UniRE on one of the supported datasets.
- Run the provided ACE2005 inference example with a trained best_model.
- Use the released ACE2005 pretrained model as a research starting point.
Requirements & limits
The repository is public and licensed under MIT. The pretrained BERT-based ACE2005 model is listed as 420 MB and linked from the README.
- The documented software versions are Python 3.7.6, PyTorch 1.8.1 and Transformers 4.2.2, plus ConfigArgParse, bidict and Fire.
- The README reports that the default training setup requires a 32 GB GPU; it suggests reducing train_batch_size and increasing gradient_accumulation_steps if memory runs out.
- The released model is for ACE2005, and its reported performance differs slightly from the paper because it was trained on a GeForce RTX 2080 Ti rather than a Quadro RTX 8000.
- The repository reports pretrained-model test results of 89.03% entity precision, 88.81% entity recall and 88.92% entity F1; strict relation precision, recall and F1 are 68.71%, 60.25% and 64.21%.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Run the documented ACE2005 inference example
Apply a trained ACE2005 UniRE model to the ACE2005 test data using the repository's inference command.
Before you start
- Python 3.7.6, PyTorch 1.8.1, Transformers 4.2.2, ConfigArgParse 1.2.3, bidict 0.20.0 and Fire 0.3.1.
- ACE2005 data prepared in the repository's expected format.
- A trained best_model in the specified save directory.
- A configured device; the README's example uses device index 0.
Steps
Prepare the ACE2005 data using the repository's data instructions.
Ensure the save directory contains the trained best_model.
Run the README's entity_relation_joint_decoder.py inference example with config.yml, the ACE2005 data directory, the save directory, device index, test flag and log file.
Expected outcome
The example runs the model's ACE2005 test inference and writes the specified test.log file.
Guide limitations
- The README documents an ACE2005 example; it does not provide the same inference example for ACE2004 or SciERC.
- The guide uses a trained best_model; the README does not describe inference without one.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 17 October 2026
- Original website title
- Receiling/UniRE · Official homepage title · retained in its source language