Today's Fine-tuning & Training: Fastest-Growing Projects — September 30, 2026
Today's Fine-tuning & Training space saw a mix of projects ranging from cybersecurity-focused AI models to advanced decision-making engines and machine learning model optimizations. Among these, JoasASantos/Offensive-Security-AI-Models stands out with its high growth score, indicating substantial interest in uncensored and fine-tuned AI models for cybersecurity tasks.
JoasASantos/Offensive-Security-AI-Models is a repository that houses uncensored AI models or those fine-tuned specifically for cybersecurity-related tasks. With a robust growth score of 95.17, it has attracted significant attention, reflected in its 412 stars and consistent development activity over the past month.
Mapika/decider offers one-pass typed decisions with calibrated probabilities, fine-tuned from Qwen3.5-2B, providing a System One style model for decision-making processes. This project's steady growth score of 57.46, alongside its substantial 919 stars and high commit activity (93 commits in the last month), suggests active engagement and development by a community interested in advanced decision algorithms.
Liuziyu77/Valen aims to train Jev-like multimodal models with vision capabilities, allowing users to create their own systems akin to System One but with enhanced visual processing. Its growth score of 51.64 and 507 stars indicate growing interest from developers looking to integrate advanced multimodal functionalities into their projects.
AkashPriyadarshii/jev-curate is a high-throughput dataset sifter for TypeSafe Jev, utilizing Rust streaming core technology with Parquet and JSONL I/O capabilities. Despite having fewer stars (91), the project's growth score of 22.67 highlights its importance in managing large-scale data preprocessing tasks efficiently.
firelex/jeff provides fine-tuned versions of Qwen3.5 and Gemma 4 for zero-shot classification purposes, aiming to enhance these models' capabilities without requiring additional training datasets. With a steady growth score of 16.91 and an impressive 999 stars, this project has garnered significant interest from developers seeking optimized AI solutions for immediate deployment.
friday-memory/friday introduces a persistent cognitive memory layer designed to improve the performance of AI coding agents like Claude & Copilot by preventing them from having "amnesia." Its growth score of 15.39 and modest star count (65) suggest it is gaining traction among developers who value continuity in their code generation tasks.
davefano/omarchy-trackpad-plus focuses on fine-grained, per-device trackpad controls and pointer-feel tuning for Omarchy, enhancing user experience through precise customization options. Although its growth score of 10.83 is relatively low, the project's 126 stars indicate a dedicated community interested in refining input device configurations.
luigisaetta/llm-fine-tuning-on-mac offers comprehensive code resources to fine-tune small language models using LoRA techniques directly on Mac devices. With a growth score of 5.17 and limited star count (21), it caters to developers seeking localized solutions for model optimization.
intikhab49/open-jev-typed-decision-engine reproduces TypeSafe Jev, providing a typed decision engine capable of handling noul/choice/score judgments efficiently. Its growth score of 3.68 and modest star count (45) reflect its niche appeal among researchers and developers focused on high-performance decision-making systems.
Zhengsw03/RL-Token-Pi05-open develops an RL Token pipeline for fine-tuning, training critical-phase classifiers, and deploying online reinforcement learning models. Despite a low growth score of 3.38 and fewer stars (36), the project's focus on specialized machine learning techniques indicates its relevance in advanced AI research circles.
Today's report highlights a diverse array of projects that cater to various aspects of model fine-tuning and training, from cybersecurity applications to decision-making engines and reinforcement learning pipelines, showcasing the dynamic nature of current AI development trends.
JoasASantos/Offensive-Security-AI-Models is a repository that houses uncensored AI models or those fine-tuned specifically for cybersecurity-related tasks. With a robust growth score of 95.17, it has attracted significant attention, reflected in its 412 stars and consistent development activity over the past month.
Mapika/decider offers one-pass typed decisions with calibrated probabilities, fine-tuned from Qwen3.5-2B, providing a System One style model for decision-making processes. This project's steady growth score of 57.46, alongside its substantial 919 stars and high commit activity (93 commits in the last month), suggests active engagement and development by a community interested in advanced decision algorithms.
Liuziyu77/Valen aims to train Jev-like multimodal models with vision capabilities, allowing users to create their own systems akin to System One but with enhanced visual processing. Its growth score of 51.64 and 507 stars indicate growing interest from developers looking to integrate advanced multimodal functionalities into their projects.
AkashPriyadarshii/jev-curate is a high-throughput dataset sifter for TypeSafe Jev, utilizing Rust streaming core technology with Parquet and JSONL I/O capabilities. Despite having fewer stars (91), the project's growth score of 22.67 highlights its importance in managing large-scale data preprocessing tasks efficiently.
firelex/jeff provides fine-tuned versions of Qwen3.5 and Gemma 4 for zero-shot classification purposes, aiming to enhance these models' capabilities without requiring additional training datasets. With a steady growth score of 16.91 and an impressive 999 stars, this project has garnered significant interest from developers seeking optimized AI solutions for immediate deployment.
friday-memory/friday introduces a persistent cognitive memory layer designed to improve the performance of AI coding agents like Claude & Copilot by preventing them from having "amnesia." Its growth score of 15.39 and modest star count (65) suggest it is gaining traction among developers who value continuity in their code generation tasks.
davefano/omarchy-trackpad-plus focuses on fine-grained, per-device trackpad controls and pointer-feel tuning for Omarchy, enhancing user experience through precise customization options. Although its growth score of 10.83 is relatively low, the project's 126 stars indicate a dedicated community interested in refining input device configurations.
luigisaetta/llm-fine-tuning-on-mac offers comprehensive code resources to fine-tune small language models using LoRA techniques directly on Mac devices. With a growth score of 5.17 and limited star count (21), it caters to developers seeking localized solutions for model optimization.
intikhab49/open-jev-typed-decision-engine reproduces TypeSafe Jev, providing a typed decision engine capable of handling noul/choice/score judgments efficiently. Its growth score of 3.68 and modest star count (45) reflect its niche appeal among researchers and developers focused on high-performance decision-making systems.
Zhengsw03/RL-Token-Pi05-open develops an RL Token pipeline for fine-tuning, training critical-phase classifiers, and deploying online reinforcement learning models. Despite a low growth score of 3.38 and fewer stars (36), the project's focus on specialized machine learning techniques indicates its relevance in advanced AI research circles.
Today's report highlights a diverse array of projects that cater to various aspects of model fine-tuning and training, from cybersecurity applications to decision-making engines and reinforcement learning pipelines, showcasing the dynamic nature of current AI development trends.