Today's Fine-tuning & Training: Fastest-Growing Projects — April 24, 2026
Today's the Fine-tuning & Training space, we saw a surge in repositories focused on multimodal fine-tuning and novel training methods. The trend suggests that developers are increasingly interested in pushing the boundaries of language models to tackle complex tasks and applications. Notably, several repositories showcasing innovative approaches to fine-tuning large language models (LLMs) have gained significant traction.
mattmireles/gemma-tuner-multimodal has taken the lead with a growth score of 67.00 and 1,384 stars. This repository provides a method for fine-tuning Gemma 4 and 3n with audio, images, and text on Apple Silicon using PyTorch and Metal Performance Shaders. Its high growth rate can be attributed to its unique approach to multimodal learning, which has garnered significant interest from the developer community.
QingGo/engram-peft boasts a growth score of 26.92 and 31 stars. This unofficial implementation of DeepSeek Engram injects high-capacity conditional memory into LLMs via sparse retrieval PEFT without increasing inference FLOPs. Its growing popularity stems from its innovative approach to enhancing the capabilities of large language models.
ZJU-OmniAI/GFT has a growth score of 18.75 and 27 stars. This repository introduces GFT, which enables imitation-to-reward fine-tuning with unbiased group advantages and dynamic coefficient rectification. Its increasing popularity can be attributed to its novel approach to training methods, which has sparked interest among researchers and developers.
WillowHe/EvoOpt_oppangu_optimization_model boasts a growth score of 10.70 and an impressive 514 stars. This repository provides solutions leveraging Openpangu - 7B as the base model for fine-tuning and application of LLMs in operations research optimization tasks. Its significant following can be attributed to its practical applications in real-world problems.
SUM-INNOVATION/RUMUS has a growth score of 9.65 and 148 stars. This Rust-based framework enables training neural networks, offering an efficient alternative to traditional methods. Its growing popularity stems from the increasing adoption of Rust as a programming language for AI applications.
semidark/kokoro-deutsch features a growth score of 8.43 and 29 stars. This project provides a complete training recipe for fine-tuning Kokoro-82M on German, offering a valuable resource for developers working with LLMs. Its growing interest can be attributed to the expanding use cases of multilingual language models.
Dynamis-Labs/spectralquant boasts a growth score of 4.95 and 125 stars. This repository introduces a method to break TurboQuant's compression limit via spectral structure, offering insights into efficient model training methods. Although its growth rate is slower compared to other repositories on this list, it remains an interesting contribution to the field.
The remaining tools, such as Mintzs/oogaboogalm, PentesterFlow/OffensiveSET, and Goekdeniz-Guelmez/moshi-finetune-mlx, have relatively lower growth scores but still demonstrate innovative approaches to fine-tuning and training.
mattmireles/gemma-tuner-multimodal has taken the lead with a growth score of 67.00 and 1,384 stars. This repository provides a method for fine-tuning Gemma 4 and 3n with audio, images, and text on Apple Silicon using PyTorch and Metal Performance Shaders. Its high growth rate can be attributed to its unique approach to multimodal learning, which has garnered significant interest from the developer community.
QingGo/engram-peft boasts a growth score of 26.92 and 31 stars. This unofficial implementation of DeepSeek Engram injects high-capacity conditional memory into LLMs via sparse retrieval PEFT without increasing inference FLOPs. Its growing popularity stems from its innovative approach to enhancing the capabilities of large language models.
ZJU-OmniAI/GFT has a growth score of 18.75 and 27 stars. This repository introduces GFT, which enables imitation-to-reward fine-tuning with unbiased group advantages and dynamic coefficient rectification. Its increasing popularity can be attributed to its novel approach to training methods, which has sparked interest among researchers and developers.
WillowHe/EvoOpt_oppangu_optimization_model boasts a growth score of 10.70 and an impressive 514 stars. This repository provides solutions leveraging Openpangu - 7B as the base model for fine-tuning and application of LLMs in operations research optimization tasks. Its significant following can be attributed to its practical applications in real-world problems.
SUM-INNOVATION/RUMUS has a growth score of 9.65 and 148 stars. This Rust-based framework enables training neural networks, offering an efficient alternative to traditional methods. Its growing popularity stems from the increasing adoption of Rust as a programming language for AI applications.
semidark/kokoro-deutsch features a growth score of 8.43 and 29 stars. This project provides a complete training recipe for fine-tuning Kokoro-82M on German, offering a valuable resource for developers working with LLMs. Its growing interest can be attributed to the expanding use cases of multilingual language models.
Dynamis-Labs/spectralquant boasts a growth score of 4.95 and 125 stars. This repository introduces a method to break TurboQuant's compression limit via spectral structure, offering insights into efficient model training methods. Although its growth rate is slower compared to other repositories on this list, it remains an interesting contribution to the field.
The remaining tools, such as Mintzs/oogaboogalm, PentesterFlow/OffensiveSET, and Goekdeniz-Guelmez/moshi-finetune-mlx, have relatively lower growth scores but still demonstrate innovative approaches to fine-tuning and training.