Today's Fine-tuning & Training: Fastest-Growing Projects — November 04, 2026
Today's the Fine-tuning & Training space on GitHub, we see a mix of projects focused on various aspects of decision-making and reinforcement learning, with some making significant strides in terms of growth and engagement from the developer community. The Juneo-Lab/decision-forge project stands out as one of the most active repositories in this category, demonstrating strong development activity over the past month.
Juneo-Lab's decision-forge compiles organizational policy into verifiable decision training data, aiding in creating robust decision-making models for enterprises and organizations. With a high growth score of 73.50 and accumulating 63 stars, it suggests that developers are finding this tool valuable for integrating structured decision-making processes within their machine learning workflows.
TuotuoLyu's FPO repository focuses on reinforcement fine-tuning of flow-matching policies specifically for vision-language-action models, aiming to optimize the performance of these complex multi-modal systems. Despite having fewer recent commits (1 in 30 days), it maintains a steady growth score and has garnered 129 stars, indicating sustained interest from researchers and practitioners working on advanced reinforcement learning applications.
FrontiersMindAI's decisionformer leverages transformers for developing decision models that can be fine-tuned, loaded, and served using any Hugging Face model. This project has seen consistent development activity with 37 commits in the last month and a moderate growth score of 10.69 alongside 44 stars, suggesting it is gaining traction among those interested in transformer-based decision-making frameworks.
Jwuthri-cloud's SelfJev offers a self-hosted decision model (selfjev-4b) with a JEV-compatible API that includes features like Python SDKs and fine-tuning capabilities. With only six commits over the last month, it still maintains a respectable growth score of 5.79 and has attracted 60 stars, indicating its potential appeal to developers seeking customizable decision-making solutions.
Sunovivid's correction-training project implements techniques for improving diffusion models through correction training methods designed to enhance their performance and stability. The repository is actively developed with 31 commits in the last month and a growth score of 5.58, alongside collecting 92 stars, reflecting its growing importance in the fine-tuning community.
Google DeepMind's hope repository introduces Hilbert Operator for Progressive Encoding (HOPE) along with Dispersed Elastic Fine-Tuning (DEFT), offering innovative approaches to model fine-tuning and optimization. With only three commits in the last month but a growth score of 3.31, it has attracted 26 stars, suggesting that while development is slower, there remains steady interest from those engaged with advanced model training techniques.
Mahmoodlab's LIMIT project focuses on vision-language fine-tuning using pathology reports, aiming to enhance medical image analysis and diagnostic capabilities through multi-modal learning approaches. It has received 22 stars and a low growth score of 1.26, indicating ongoing but less pronounced activity compared to other projects in the category.
Shrikant922's training-data-bot automates the process from raw documents to production fine-tuning datasets for large language models (LLMs), streamlining data preparation workflows. With just one commit over the last month and a minimal growth score of 0.91, it has still managed to attract 47 stars, hinting at its utility in specific use cases despite lower recent activity levels.
Overall, these repositories showcase diverse approaches to fine-tuning and training models across various domains, from enterprise decision-making to advanced reinforcement learning applications, each contributing unique value based on their distinct features and active development.
Juneo-Lab's decision-forge compiles organizational policy into verifiable decision training data, aiding in creating robust decision-making models for enterprises and organizations. With a high growth score of 73.50 and accumulating 63 stars, it suggests that developers are finding this tool valuable for integrating structured decision-making processes within their machine learning workflows.
TuotuoLyu's FPO repository focuses on reinforcement fine-tuning of flow-matching policies specifically for vision-language-action models, aiming to optimize the performance of these complex multi-modal systems. Despite having fewer recent commits (1 in 30 days), it maintains a steady growth score and has garnered 129 stars, indicating sustained interest from researchers and practitioners working on advanced reinforcement learning applications.
FrontiersMindAI's decisionformer leverages transformers for developing decision models that can be fine-tuned, loaded, and served using any Hugging Face model. This project has seen consistent development activity with 37 commits in the last month and a moderate growth score of 10.69 alongside 44 stars, suggesting it is gaining traction among those interested in transformer-based decision-making frameworks.
Jwuthri-cloud's SelfJev offers a self-hosted decision model (selfjev-4b) with a JEV-compatible API that includes features like Python SDKs and fine-tuning capabilities. With only six commits over the last month, it still maintains a respectable growth score of 5.79 and has attracted 60 stars, indicating its potential appeal to developers seeking customizable decision-making solutions.
Sunovivid's correction-training project implements techniques for improving diffusion models through correction training methods designed to enhance their performance and stability. The repository is actively developed with 31 commits in the last month and a growth score of 5.58, alongside collecting 92 stars, reflecting its growing importance in the fine-tuning community.
Google DeepMind's hope repository introduces Hilbert Operator for Progressive Encoding (HOPE) along with Dispersed Elastic Fine-Tuning (DEFT), offering innovative approaches to model fine-tuning and optimization. With only three commits in the last month but a growth score of 3.31, it has attracted 26 stars, suggesting that while development is slower, there remains steady interest from those engaged with advanced model training techniques.
Mahmoodlab's LIMIT project focuses on vision-language fine-tuning using pathology reports, aiming to enhance medical image analysis and diagnostic capabilities through multi-modal learning approaches. It has received 22 stars and a low growth score of 1.26, indicating ongoing but less pronounced activity compared to other projects in the category.
Shrikant922's training-data-bot automates the process from raw documents to production fine-tuning datasets for large language models (LLMs), streamlining data preparation workflows. With just one commit over the last month and a minimal growth score of 0.91, it has still managed to attract 47 stars, hinting at its utility in specific use cases despite lower recent activity levels.
Overall, these repositories showcase diverse approaches to fine-tuning and training models across various domains, from enterprise decision-making to advanced reinforcement learning applications, each contributing unique value based on their distinct features and active development.