Today's Fine-tuning & Training: Fastest-Growing Projects — September 19, 2026
Today's fine-tuning and training landscape on GitHub continues to showcase a diverse range of projects catering to various aspects of AI model development, from specialized decision-making models to comprehensive frameworks for world action modeling. These tools reflect the growing demand for customizable and efficient solutions that can be tailored to specific tasks or environments.
Mapika/decider is a one-pass typed decision engine fine-tuned from Qwen3.5-2B, designed to provide calibrated probability estimates akin to System One style models. With its growth score of 43.17 and an increasing number of stars (currently at 25), this project stands out for its unique approach to decision-making processes in AI systems.
davefano/omarchy-trackpad-plus offers fine-grained control over trackpad settings and pointer feel tailored specifically for the Omarchy environment, demonstrating a niche but vital application of tuning technologies. Its high growth score (33.71) and substantial star count (121) suggest strong community interest in enhancing user experience through precise device customization.
friday-memory/friday introduces an open-source cognitive memory layer aimed at ensuring persistent functionality for AI coding assistants, addressing the issue of 'amnesia' commonly experienced by these tools. With a growth score of 19.33 and 38 stars, this project garners attention for its potential to significantly improve the reliability and continuity of intelligent programming aids.
OpenMOSS/EasyWAM presents a unified framework that simplifies training, fine-tuning, and evaluation processes for World Action Models (WAMs). This comprehensive tool has attracted considerable interest, as evidenced by its growth score of 16.40 and 295 stars, highlighting the community's need for streamlined methodologies in WAM development.
Junchao-cs/SolarWM focuses on providing open data and scalable training methods for long-horizon video world models, a critical area for advancing AI capabilities in predictive analytics and visual understanding. With an impressive growth score of 14.12 and 678 stars, this project is making waves due to its potential impact on the future of computer vision and machine learning research.
njgymb/diffusers-sculptor introduces a lightweight, configuration-driven pipeline for efficient fine-tuning of Stable Diffusion models. Its growth score of 13.12 coupled with 56 stars indicates that developers are increasingly seeking modular tools to enhance their AI model customization workflows efficiently.
Similarly, R0650N/spirit-flux-refiner and nrodriguez1997/lora-docker-aliyun-pipeline both achieve a growth score of 13.10, each offering specialized solutions for Flux1-LoRA training and automated LoRA training pipelines on Alibaba Cloud respectively. These projects attract attention due to their targeted approach towards simplifying complex AI model fine-tuning tasks.
The KingHsp/vram-sage-training suite is designed specifically for SDXL and ANIMA models, optimizing them for 12GB GPUs through VRAM-smart techniques. This project's growth score of 13.10 alongside its growing star count (55) highlights the ongoing demand for tools that enable high-performance model training on resource-constrained hardware.
Lastly, alrisqi/MrD-Optimizer-Lab provides a suite of AI training and optimization tools under the MrD Deep Learning Studio umbrella, aiming to simplify complex development processes. With a growth score of 13.10 and 55 stars, this project is gaining traction among developers looking for comprehensive solutions that streamline their model training workflows.
These projects collectively underscore the evolving nature of fine-tuning and training tools in the AI domain, reflecting both broad and specialized needs within the developer community.
Mapika/decider is a one-pass typed decision engine fine-tuned from Qwen3.5-2B, designed to provide calibrated probability estimates akin to System One style models. With its growth score of 43.17 and an increasing number of stars (currently at 25), this project stands out for its unique approach to decision-making processes in AI systems.
davefano/omarchy-trackpad-plus offers fine-grained control over trackpad settings and pointer feel tailored specifically for the Omarchy environment, demonstrating a niche but vital application of tuning technologies. Its high growth score (33.71) and substantial star count (121) suggest strong community interest in enhancing user experience through precise device customization.
friday-memory/friday introduces an open-source cognitive memory layer aimed at ensuring persistent functionality for AI coding assistants, addressing the issue of 'amnesia' commonly experienced by these tools. With a growth score of 19.33 and 38 stars, this project garners attention for its potential to significantly improve the reliability and continuity of intelligent programming aids.
OpenMOSS/EasyWAM presents a unified framework that simplifies training, fine-tuning, and evaluation processes for World Action Models (WAMs). This comprehensive tool has attracted considerable interest, as evidenced by its growth score of 16.40 and 295 stars, highlighting the community's need for streamlined methodologies in WAM development.
Junchao-cs/SolarWM focuses on providing open data and scalable training methods for long-horizon video world models, a critical area for advancing AI capabilities in predictive analytics and visual understanding. With an impressive growth score of 14.12 and 678 stars, this project is making waves due to its potential impact on the future of computer vision and machine learning research.
njgymb/diffusers-sculptor introduces a lightweight, configuration-driven pipeline for efficient fine-tuning of Stable Diffusion models. Its growth score of 13.12 coupled with 56 stars indicates that developers are increasingly seeking modular tools to enhance their AI model customization workflows efficiently.
Similarly, R0650N/spirit-flux-refiner and nrodriguez1997/lora-docker-aliyun-pipeline both achieve a growth score of 13.10, each offering specialized solutions for Flux1-LoRA training and automated LoRA training pipelines on Alibaba Cloud respectively. These projects attract attention due to their targeted approach towards simplifying complex AI model fine-tuning tasks.
The KingHsp/vram-sage-training suite is designed specifically for SDXL and ANIMA models, optimizing them for 12GB GPUs through VRAM-smart techniques. This project's growth score of 13.10 alongside its growing star count (55) highlights the ongoing demand for tools that enable high-performance model training on resource-constrained hardware.
Lastly, alrisqi/MrD-Optimizer-Lab provides a suite of AI training and optimization tools under the MrD Deep Learning Studio umbrella, aiming to simplify complex development processes. With a growth score of 13.10 and 55 stars, this project is gaining traction among developers looking for comprehensive solutions that streamline their model training workflows.
These projects collectively underscore the evolving nature of fine-tuning and training tools in the AI domain, reflecting both broad and specialized needs within the developer community.