Today's Fine-tuning & Training: Fastest-Growing Projects — October 02, 2026
Today's radar highlights a notable surge in repositories focusing on fine-tuning and training AI models for specialized tasks such as cybersecurity and decision-making systems. Among these, JoasASantos/Offensive-Security-AI-Models stands out with its uncensored approach to AI models tailored for offensive security purposes, accumulating 518 stars and showing a robust growth score of 73.70. The repository Mapika/decider is also making waves with its system designed to make decisions based on calibrated probabilities, drawing significant interest with over 1,029 stars.
JoasASantos/Offensive-Security-AI-Models offers uncensored AI models or those fine-tuned for cybersecurity tasks, aiming to provide tools that can be used in offensive security scenarios. Its high growth score and substantial star count suggest a strong community engagement and practical application interest from security professionals and researchers alike.
Mapika/decider provides a one-pass typed decision system with calibrated probabilities, fine-tuned from Qwen3.5-2B. The repository's extensive recent activity (94 commits in the last month) alongside its impressive star count reflects its relevance and utility for developers working on decision-making systems that require high accuracy and efficiency.
Liuziyu77/Valen allows users to train a Jev-like multimodal model, enhancing the System One Model with visual capabilities. With 537 stars and a growth score of 41.83, Valen demonstrates growing interest from researchers and developers seeking to integrate vision into decision-making models.
AkashPriyadarshii/jev-curate is designed for high-throughput synthetic dataset sifting, supporting TypeSafe Jev with Rust streaming core capabilities. Despite its lower growth score of 19.89, the repository's significant number of commits (70 in the last month) and 95 stars indicate active development and community support.
friday-memory/friday offers a persistent cognitive memory layer for AI coding agents to prevent amnesia issues, aiming to enhance the performance of tools like Claude & Copilot. With 66 stars and a growth score of 14.06, this project shows steady interest from developers looking to improve continuous learning capabilities in AI-assisted coding environments.
davefano/omarchy-trackpad-plus focuses on fine-grained trackpad controls for Omarchy devices, enabling precise adjustments through software tuning. With 39 commits in the last month and 126 stars, this project reflects a niche but active community interested in enhancing user experience through detailed input device control.
intikhab49/open-jev-typed-decision-engine is an open-source reproduction of TypeSafe Jev, offering a decision engine that operates with calibrated confidence. Its modest growth score (3.08) and 44 stars suggest it has found a dedicated audience interested in replicating advanced decision-making systems.
Zhengsw03/RL-Token-Pi05-open provides an RL Token pipeline for fine-tuning the SO-101 arm, incorporating critical-phase classifiers and online reinforcement learning techniques. With 42 stars and a growth score of 3.00, this repository shows steady development in the field of reinforcement learning applied to AI model training.
Thinking-Space/One-Shot-OPD rethinks on-policy distillation for large language models with an emphasis on efficient training using just one example. Its growth score (2.26) and 92 stars indicate a growing interest among researchers looking to optimize the training process of large-scale AI models.
These repositories collectively showcase the diversity and depth of ongoing advancements in fine-tuning and training AI systems, from specialized cybersecurity tools to sophisticated decision-making engines and memory layers for continuous learning. The high levels of activity and community engagement across these projects underscore the dynamic nature of current AI research and development efforts.
JoasASantos/Offensive-Security-AI-Models offers uncensored AI models or those fine-tuned for cybersecurity tasks, aiming to provide tools that can be used in offensive security scenarios. Its high growth score and substantial star count suggest a strong community engagement and practical application interest from security professionals and researchers alike.
Mapika/decider provides a one-pass typed decision system with calibrated probabilities, fine-tuned from Qwen3.5-2B. The repository's extensive recent activity (94 commits in the last month) alongside its impressive star count reflects its relevance and utility for developers working on decision-making systems that require high accuracy and efficiency.
Liuziyu77/Valen allows users to train a Jev-like multimodal model, enhancing the System One Model with visual capabilities. With 537 stars and a growth score of 41.83, Valen demonstrates growing interest from researchers and developers seeking to integrate vision into decision-making models.
AkashPriyadarshii/jev-curate is designed for high-throughput synthetic dataset sifting, supporting TypeSafe Jev with Rust streaming core capabilities. Despite its lower growth score of 19.89, the repository's significant number of commits (70 in the last month) and 95 stars indicate active development and community support.
friday-memory/friday offers a persistent cognitive memory layer for AI coding agents to prevent amnesia issues, aiming to enhance the performance of tools like Claude & Copilot. With 66 stars and a growth score of 14.06, this project shows steady interest from developers looking to improve continuous learning capabilities in AI-assisted coding environments.
davefano/omarchy-trackpad-plus focuses on fine-grained trackpad controls for Omarchy devices, enabling precise adjustments through software tuning. With 39 commits in the last month and 126 stars, this project reflects a niche but active community interested in enhancing user experience through detailed input device control.
intikhab49/open-jev-typed-decision-engine is an open-source reproduction of TypeSafe Jev, offering a decision engine that operates with calibrated confidence. Its modest growth score (3.08) and 44 stars suggest it has found a dedicated audience interested in replicating advanced decision-making systems.
Zhengsw03/RL-Token-Pi05-open provides an RL Token pipeline for fine-tuning the SO-101 arm, incorporating critical-phase classifiers and online reinforcement learning techniques. With 42 stars and a growth score of 3.00, this repository shows steady development in the field of reinforcement learning applied to AI model training.
Thinking-Space/One-Shot-OPD rethinks on-policy distillation for large language models with an emphasis on efficient training using just one example. Its growth score (2.26) and 92 stars indicate a growing interest among researchers looking to optimize the training process of large-scale AI models.
These repositories collectively showcase the diversity and depth of ongoing advancements in fine-tuning and training AI systems, from specialized cybersecurity tools to sophisticated decision-making engines and memory layers for continuous learning. The high levels of activity and community engagement across these projects underscore the dynamic nature of current AI research and development efforts.