Today's Fine-tuning & Training: Fastest-Growing Projects — September 29, 2026
This week, the Fine-tuning & Training category on GitHub continues to showcase a diverse range of projects that span various applications and industries, from cybersecurity to decision-making models. The trend highlights the increasing importance of fine-tuning AI models for specific tasks or domains, demonstrating the growing demand for customized solutions in the realm of artificial intelligence.
The project JoasASantos/Offensive-Security-AI-Models offers uncensored AI models or those fine-tuned specifically for cybersecurity tasks. With a growth score of 64.75 and accumulating 169 stars, this repository stands out due to its unique approach in providing security-relevant AI models that can be tailored for specific offensive measures.
Mapika/decider is a one-pass typed decision-making model fine-tuned from Qwen3.5-2B, designed to provide calibrated probabilities similar to the System One style model. The project's significant growth score of 58.77 and 880 stars indicate its popularity among developers interested in robust decision engines that can handle complex probabilistic assessments efficiently.
Liuziyu77/Valen is a Jev-like multimodal model training framework that allows users to train their own vision-enabled systems similar to System One models. With a growth score of 51.17 and 437 stars, this repository appeals to researchers and developers seeking to integrate visual data into decision-making processes.
AkashPriyadarshii/jev-curate is designed as a high-throughput dataset sifter for TypeSafe Jev, featuring Rust streaming core capabilities and efficient I/O operations. Despite having a lower growth score of 24.64, it still garners attention with 89 stars due to its specialized focus on optimizing the performance of decision engines through synthetic and pretraining datasets.
davefano/omarchy-trackpad-plus focuses on fine-grained control over trackpad settings for Omarchy devices, offering users customizable pointer-feel tuning options. Although its growth score is relatively low at 11.47, it has managed to attract 126 stars from developers interested in enhancing the user experience through precise device customization.
friday-memory/friday introduces a persistent cognitive memory layer designed specifically for AI coding agents, ensuring that these assistants retain context and information across sessions. With a growth score of 10.92 and 62 stars, this repository addresses an essential need for continuous learning in AI-driven software development tools.
luigisaetta/llm-fine-tuning-on-mac provides detailed code to fine-tune small language models using LoRA techniques on macOS systems. Its modest growth score of 5.36 alongside 21 stars suggests a niche audience interested in local machine learning model customization and performance optimization.
intikhab49/open-jev-typed-decision-engine offers an open-source reproduction of TypeSafe Jev, aiming to provide a faster and more calibrated decision engine compared to its original counterpart. With a growth score of 4.05 and 45 stars, this project attracts those looking for efficient alternatives in the realm of typed decision-making systems.
firelex/jeff focuses on fine-tuning Qwen3.5 and Gemma 4 models for zero-shot classification tasks. Despite its relatively lower growth score of 3.53 and 158 stars, it appeals to developers interested in leveraging pre-trained models for immediate deployment without extensive retraining phases.
Thinking-Space/One-Shot-OPD presents a novel approach to on-policy distillation in large language models with an emphasis on efficiency through one-shot training examples. With a growth score of 2.40 and 86 stars, this project resonates with researchers exploring innovative methods for enhancing model performance while minimizing computational overhead.
These projects collectively illustrate the breadth and depth of fine-tuning and training methodologies being explored across different domains, underscoring the dynamic nature of AI development today.
The project JoasASantos/Offensive-Security-AI-Models offers uncensored AI models or those fine-tuned specifically for cybersecurity tasks. With a growth score of 64.75 and accumulating 169 stars, this repository stands out due to its unique approach in providing security-relevant AI models that can be tailored for specific offensive measures.
Mapika/decider is a one-pass typed decision-making model fine-tuned from Qwen3.5-2B, designed to provide calibrated probabilities similar to the System One style model. The project's significant growth score of 58.77 and 880 stars indicate its popularity among developers interested in robust decision engines that can handle complex probabilistic assessments efficiently.
Liuziyu77/Valen is a Jev-like multimodal model training framework that allows users to train their own vision-enabled systems similar to System One models. With a growth score of 51.17 and 437 stars, this repository appeals to researchers and developers seeking to integrate visual data into decision-making processes.
AkashPriyadarshii/jev-curate is designed as a high-throughput dataset sifter for TypeSafe Jev, featuring Rust streaming core capabilities and efficient I/O operations. Despite having a lower growth score of 24.64, it still garners attention with 89 stars due to its specialized focus on optimizing the performance of decision engines through synthetic and pretraining datasets.
davefano/omarchy-trackpad-plus focuses on fine-grained control over trackpad settings for Omarchy devices, offering users customizable pointer-feel tuning options. Although its growth score is relatively low at 11.47, it has managed to attract 126 stars from developers interested in enhancing the user experience through precise device customization.
friday-memory/friday introduces a persistent cognitive memory layer designed specifically for AI coding agents, ensuring that these assistants retain context and information across sessions. With a growth score of 10.92 and 62 stars, this repository addresses an essential need for continuous learning in AI-driven software development tools.
luigisaetta/llm-fine-tuning-on-mac provides detailed code to fine-tune small language models using LoRA techniques on macOS systems. Its modest growth score of 5.36 alongside 21 stars suggests a niche audience interested in local machine learning model customization and performance optimization.
intikhab49/open-jev-typed-decision-engine offers an open-source reproduction of TypeSafe Jev, aiming to provide a faster and more calibrated decision engine compared to its original counterpart. With a growth score of 4.05 and 45 stars, this project attracts those looking for efficient alternatives in the realm of typed decision-making systems.
firelex/jeff focuses on fine-tuning Qwen3.5 and Gemma 4 models for zero-shot classification tasks. Despite its relatively lower growth score of 3.53 and 158 stars, it appeals to developers interested in leveraging pre-trained models for immediate deployment without extensive retraining phases.
Thinking-Space/One-Shot-OPD presents a novel approach to on-policy distillation in large language models with an emphasis on efficiency through one-shot training examples. With a growth score of 2.40 and 86 stars, this project resonates with researchers exploring innovative methods for enhancing model performance while minimizing computational overhead.
These projects collectively illustrate the breadth and depth of fine-tuning and training methodologies being explored across different domains, underscoring the dynamic nature of AI development today.