Today's Fine-tuning & Training: Fastest-Growing Projects — October 25, 2026
This week, the Fine-tuning & Training space on GitHub continues to see significant activity as developers and researchers push the boundaries of AI model customization and performance optimization. Among the growing repositories this week, jwuthri-cloud/SelfJev stands out with its high growth score, reflecting substantial interest in self-hosted decision models.
jwuthri-cloud/SelfJev is a repository that offers a self-hosted decision model (selfjev-4b) compatible with JEV API standards. It includes tools for fine-tuning and reinforcement learning-based continuous deployment, making it appealing to developers looking to customize their AI decision-making processes without relying on third-party services. The high growth score of 56.12 and 53 stars indicate a growing community interested in deploying self-hosted models tailored to specific needs.
FrontiersMindAI/decisionformer provides a framework for fine-tuning, loading, and serving calibrated decision models using Hugging Face transformers. This repository is designed to streamline the process of integrating decision-making capabilities into existing machine learning workflows, making it easier for developers to enhance their applications with sophisticated AI-driven decisions. With 36 in growth score and 30 stars, this project is gaining traction among those looking to leverage pre-trained models for decision tasks without extensive customization.
JoasASantos/Offensive-Security-AI-Models offers a collection of uncensored or fine-tuned AI models specifically targeted at cybersecurity applications. The repository's focus on security-related tasks sets it apart and attracts developers and researchers interested in these specialized use cases. With 729 stars, the project reflects strong community interest in leveraging AI for advanced threat detection and response mechanisms.
AcademiaSD/AcademiaSD_LoRAlab-TrainerStudio is a training studio that simplifies the process of fine-tuning LoRAs (low-rank adaptation) across various large language models like Qwen-Image 2.1, FLUX.2 Klein 9B, and more. The platform supports NVIDIA GPUs with varying VRAM capacities and caters to Windows, Linux, and RunPod environments. This repository's growth score of 13.91 and 112 stars suggest it is becoming a popular tool for researchers and developers looking to fine-tune models without the need for extensive computational resources.
Baimuking/AI-Jailbreak compiles a comprehensive collection of AI jailbreaking techniques across different model types, including GPT, Claude, Gemini, DeepSeek, GLM, Kimi, Grok, Qwen, and local models. The repository aims to provide一键安装的破解卡覆盖多种模型，使其成为对AI限制措施感兴趣的开发者的宝贵资源。该项目的增长得分为13.72且拥有118颗星，表明其在关注AI限制和自由度问题的技术社区中具有重要地位。
sunovivid/correction-training is the official implementation of "Correction Training for Diffusion Models," which focuses on improving diffusion models through correction training techniques. This repository's focus on enhancing model accuracy and stability aligns with ongoing research trends in machine learning optimization. With a growth score of 8.69 and 83 stars, it indicates growing interest from researchers aiming to refine their diffusion models for better performance.
qlabs-eng/dust introduces a novel approach to pretraining transformers without backpropagation, which could be particularly appealing to those interested in reducing computational costs during the training process. The repository's growth score of 8.59 and 263 stars reflect its appeal to researchers seeking efficient model training strategies that minimize resource consumption.
empero-org/brewery-ai is an intuitive console agent designed to simplify the fine-tuning of AI models, making it accessible even for those with minimal experience in machine learning. This tool aims to lower the barrier to entry for AI customization and experimentation by providing a user-friendly interface. The repository's growth score of 8.21 and 210 stars highlight its growing popularity among developers looking for streamlined model fine-tuning solutions.
POInfra-AI/POInfra focuses on policy optimization for generative models, incorporating techniques like diffusion and flow policies for online fine-tuning and inference-time guidance. This repository targets researchers and practitioners working with advanced generative AI systems who seek to optimize their performance through dynamic policy adjustments. With a growth score of 6.09 and 92 stars, it suggests that there is an emerging community interested in applying these techniques to improve model flexibility and adaptability.
banadmiralpurge/lora-merge, with its modest growth score of 4.50 but still notable given the date mentioned (which seems incorrect), provides tools for merging LoRA models, enabling users to combine multiple fine-tuned layers into a single cohesive model. While it has fewer stars at 27, the project's focus on simplifying the process of integrating different
jwuthri-cloud/SelfJev is a repository that offers a self-hosted decision model (selfjev-4b) compatible with JEV API standards. It includes tools for fine-tuning and reinforcement learning-based continuous deployment, making it appealing to developers looking to customize their AI decision-making processes without relying on third-party services. The high growth score of 56.12 and 53 stars indicate a growing community interested in deploying self-hosted models tailored to specific needs.
FrontiersMindAI/decisionformer provides a framework for fine-tuning, loading, and serving calibrated decision models using Hugging Face transformers. This repository is designed to streamline the process of integrating decision-making capabilities into existing machine learning workflows, making it easier for developers to enhance their applications with sophisticated AI-driven decisions. With 36 in growth score and 30 stars, this project is gaining traction among those looking to leverage pre-trained models for decision tasks without extensive customization.
JoasASantos/Offensive-Security-AI-Models offers a collection of uncensored or fine-tuned AI models specifically targeted at cybersecurity applications. The repository's focus on security-related tasks sets it apart and attracts developers and researchers interested in these specialized use cases. With 729 stars, the project reflects strong community interest in leveraging AI for advanced threat detection and response mechanisms.
AcademiaSD/AcademiaSD_LoRAlab-TrainerStudio is a training studio that simplifies the process of fine-tuning LoRAs (low-rank adaptation) across various large language models like Qwen-Image 2.1, FLUX.2 Klein 9B, and more. The platform supports NVIDIA GPUs with varying VRAM capacities and caters to Windows, Linux, and RunPod environments. This repository's growth score of 13.91 and 112 stars suggest it is becoming a popular tool for researchers and developers looking to fine-tune models without the need for extensive computational resources.
Baimuking/AI-Jailbreak compiles a comprehensive collection of AI jailbreaking techniques across different model types, including GPT, Claude, Gemini, DeepSeek, GLM, Kimi, Grok, Qwen, and local models. The repository aims to provide一键安装的破解卡覆盖多种模型，使其成为对AI限制措施感兴趣的开发者的宝贵资源。该项目的增长得分为13.72且拥有118颗星，表明其在关注AI限制和自由度问题的技术社区中具有重要地位。
sunovivid/correction-training is the official implementation of "Correction Training for Diffusion Models," which focuses on improving diffusion models through correction training techniques. This repository's focus on enhancing model accuracy and stability aligns with ongoing research trends in machine learning optimization. With a growth score of 8.69 and 83 stars, it indicates growing interest from researchers aiming to refine their diffusion models for better performance.
qlabs-eng/dust introduces a novel approach to pretraining transformers without backpropagation, which could be particularly appealing to those interested in reducing computational costs during the training process. The repository's growth score of 8.59 and 263 stars reflect its appeal to researchers seeking efficient model training strategies that minimize resource consumption.
empero-org/brewery-ai is an intuitive console agent designed to simplify the fine-tuning of AI models, making it accessible even for those with minimal experience in machine learning. This tool aims to lower the barrier to entry for AI customization and experimentation by providing a user-friendly interface. The repository's growth score of 8.21 and 210 stars highlight its growing popularity among developers looking for streamlined model fine-tuning solutions.
POInfra-AI/POInfra focuses on policy optimization for generative models, incorporating techniques like diffusion and flow policies for online fine-tuning and inference-time guidance. This repository targets researchers and practitioners working with advanced generative AI systems who seek to optimize their performance through dynamic policy adjustments. With a growth score of 6.09 and 92 stars, it suggests that there is an emerging community interested in applying these techniques to improve model flexibility and adaptability.
banadmiralpurge/lora-merge, with its modest growth score of 4.50 but still notable given the date mentioned (which seems incorrect), provides tools for merging LoRA models, enabling users to combine multiple fine-tuned layers into a single cohesive model. While it has fewer stars at 27, the project's focus on simplifying the process of integrating different