Dinothesia will hold an open-source hackathon to improve large language model (LLM) training and evaluation tools it developed in-house together with external developers. Dinothesia, a company specialising in long-term memory artificial intelligence (AI) and semiconductor integrated solutions, said on Sunday it will run two sessions on Nov. 14 and Nov. 21 at Modu Research Institute’s Gangnam site. Participants will practise running the tools, modifying the code and submitting pull requests (PRs) proposing changes.

Dinothesia overhauled internal training and evaluation tools it has used for model development so they can be used externally and released them on GitHub. The company said the aim is to share its accumulated LLM development experience and increase tool usage and external developer contributions.

The hackathon will be conducted using the released code. Dinothesia’s 'DNA Factory' is a fine-tuning tool that further trains existing models and helps users run it by configuring training methods, data and graphics processing unit (GPU) environments. The 'DNA Evaluation Harness' evaluates model-generated responses against set criteria, allowing users to check performance, response characteristics and areas for improvement.

The programme consists of an introduction to the tools, guidance on how to use GPUs, an explanation of GitHub contribution procedures and hands-on practice. Participants will run the tools in a GPU environment, modify code and draft PRs in a collaborative process to propose improvements to the project.

The event will be held in connection with the Ministry of Science and ICT and the National IT Industry Promotion Agency (NIPA) '2026 Open-Source AI and SW Development and Utilisation Support Project'. Dinothesia is promoting open-source conversion and advancement of its LLM training and evaluation tools under the project’s 'DNA Suite' task. It plans to reflect user feedback and external developer contributions gathered at the event in future tool improvements and continue exchanges with the developer community.

Lee Seung-jae (이승재), Dinothesia’s chief AI officer, said, "We aim to reduce the burden on other companies and researchers of building development environments by sharing training and evaluation tools and development experience, and help them focus on improving their own models." He added, "Using this hackathon as a starting point, we will expand external developers’ use and contributions and develop it into an open-source project that can be used and improved together in various settings."