Today's Fine-tuning & Training: Fastest-Growing Projects — May 15, 2026
This week, the Fine-tuning & Training category on GitHub continues to showcase a range of innovative solutions designed to optimize and streamline the process of fine-tuning large language models (LLMs). Among the standout projects this week is declare-lab's delta-Mem, which introduces an efficient online memory system for LLMs. Additionally, generative-computing’s Granite Switch stands out with its ability to offer high accuracy across multiple fine-tuned models while maintaining a compact footprint.
declare-lab/delta-Mem is the official repository of the paper "delta-Mem: Efficient Online Memory for Large Language Models." This project aims to enhance the efficiency and effectiveness of online memory systems used in large language model training. With a growth score of 8.43 and 64 stars, it highlights an ongoing interest in optimizing memory management within LLMs, which is crucial for performance enhancement.
generative-computing/granite-switch offers a solution that combines the accuracy benefits of multiple fine-tuned models with the efficiency of using just one model footprint. The project has seen significant activity over the past month, with 23 commits and a growth score of 7.64, indicating growing community interest in this innovative approach to reducing computational overhead while maintaining high performance.
UNfukashigi/Anima-LoRA-Factory is a user-friendly GUI tool designed for training LoRAs (Low-Rank Adaptation) for next-generation Anima diffusion models. The project’s growth score of 5.83 and 30 stars reflect the growing demand for accessible tools that simplify the process of fine-tuning LLMs, making it easier for developers to experiment with model personalization.
ZJU-OmniAI/GFT presents a framework for reward fine-tuning in imitation learning scenarios, introducing unbiased group advantages and dynamic coefficient rectification. With 45 commits over the past month and a growth score of 5.28, this project is gaining traction among researchers and developers who are interested in improving the efficiency and effectiveness of reinforcement learning techniques applied to LLMs.
Jackohhhh/MedLLM-Finetuning offers an out-of-the-box framework for fine-tuning large language models specifically tailored for medical binary classification tasks. The project has garnered 21 stars and a growth score of 2.53, reflecting its relevance in the healthcare sector where there is a growing need for specialized LLM applications that can support clinical decision-making.
raiyan Yahya's how-to-train-your-gpt repository provides an educational resource for building modern large language models from scratch, with every line commented and explained. Despite having no recent commits or stars recorded, this project remains valuable for beginners looking to understand the foundational aspects of training LLMs.
declare-lab/delta-Mem is the official repository of the paper "delta-Mem: Efficient Online Memory for Large Language Models." This project aims to enhance the efficiency and effectiveness of online memory systems used in large language model training. With a growth score of 8.43 and 64 stars, it highlights an ongoing interest in optimizing memory management within LLMs, which is crucial for performance enhancement.
generative-computing/granite-switch offers a solution that combines the accuracy benefits of multiple fine-tuned models with the efficiency of using just one model footprint. The project has seen significant activity over the past month, with 23 commits and a growth score of 7.64, indicating growing community interest in this innovative approach to reducing computational overhead while maintaining high performance.
UNfukashigi/Anima-LoRA-Factory is a user-friendly GUI tool designed for training LoRAs (Low-Rank Adaptation) for next-generation Anima diffusion models. The project’s growth score of 5.83 and 30 stars reflect the growing demand for accessible tools that simplify the process of fine-tuning LLMs, making it easier for developers to experiment with model personalization.
ZJU-OmniAI/GFT presents a framework for reward fine-tuning in imitation learning scenarios, introducing unbiased group advantages and dynamic coefficient rectification. With 45 commits over the past month and a growth score of 5.28, this project is gaining traction among researchers and developers who are interested in improving the efficiency and effectiveness of reinforcement learning techniques applied to LLMs.
Jackohhhh/MedLLM-Finetuning offers an out-of-the-box framework for fine-tuning large language models specifically tailored for medical binary classification tasks. The project has garnered 21 stars and a growth score of 2.53, reflecting its relevance in the healthcare sector where there is a growing need for specialized LLM applications that can support clinical decision-making.
raiyan Yahya's how-to-train-your-gpt repository provides an educational resource for building modern large language models from scratch, with every line commented and explained. Despite having no recent commits or stars recorded, this project remains valuable for beginners looking to understand the foundational aspects of training LLMs.