Today's Fine-tuning & Training: Fastest-Growing Projects — April 25, 2026
Today's the Fine-tuning & Training space, we're seeing a surge in innovative solutions for optimizing and fine-tuning large language models (LLMs) and other AI architectures. From multimodal fine-tuning to sparse retrieval techniques, researchers and developers are pushing the boundaries of what's possible with these powerful tools. As a result, several repositories have seen significant growth in popularity.
mattmireles/gemma-tuner-multimodal has taken the top spot this week, with a Growth Score of 63.31 and 1,385 stars. This repository provides a method for fine-tuning Gemma 4 and 3n models on Apple Silicon using PyTorch and Metal Performance Shaders, allowing for efficient training on diverse data types including audio, images, and text. Its popularity can be attributed to the growing interest in multimodal learning and the need for optimized training methods.
QingGo/engram-peft has also seen significant growth, with a Growth Score of 24.85 and 31 stars. This unofficial implementation of DeepSeek Engram enables the injection of high-capacity conditional memory into LLMs via sparse retrieval PEFT, without increasing inference FLOPs. Its growing popularity stems from the potential for improving LLM performance while maintaining efficiency.
ZJU-OmniAI/GFT has a Growth Score of 16.72 and 28 stars, and offers a novel approach to fine-tuning called GFT, which leverages unbiased group advantages and dynamic coefficient rectification. This repository is gaining traction due to its potential for improving the stability and effectiveness of reward-based fine-tuning methods.
UNfukashigi/Anima-LoRA-Factory has a Growth Score of 14.50 and 21 stars, providing a user-friendly GUI tool for training LoRAs (Low-Rank Adaptation) for next-generation Anima diffusion models. Its growth can be attributed to the increasing popularity of LoRA as a fine-tuning technique and the need for accessible tools.
WillowHe/EvoOpt_oppangu_optimization_model has a Growth Score of 10.29 and an impressive 514 stars, offering solutions for leveraging Openpangu-7B as a base model for fine-tuning and applying large language models to operations research optimization tasks. Its popularity stems from the growing interest in using LLMs for real-world applications.
SUM-INNOVATION/RUMUS has a Growth Score of 9.80 and 163 stars, providing a Rust-based framework for training neural networks. Its growth can be attributed to the increasing adoption of Rust as a programming language for AI development.
semidark/kokoro-deutsch has a Growth Score of 7.87 and 29 stars, offering a complete recipe for fine-tuning Kokoro-82M on new languages, in this case German. This repository is gaining traction due to its potential for improving multilingual support in LLMs.
Other notable mentions include Dynamis-Labs/spectralquant, which explores spectral structure to improve compression limits; Mintzs/oogaboogalm, which proposes fine-tuning AI models with caveman system prompts; and PentesterFlow/OffensiveSET, which generates high-quality pentesting conversation datasets for LLM fine-tuning.
mattmireles/gemma-tuner-multimodal has taken the top spot this week, with a Growth Score of 63.31 and 1,385 stars. This repository provides a method for fine-tuning Gemma 4 and 3n models on Apple Silicon using PyTorch and Metal Performance Shaders, allowing for efficient training on diverse data types including audio, images, and text. Its popularity can be attributed to the growing interest in multimodal learning and the need for optimized training methods.
QingGo/engram-peft has also seen significant growth, with a Growth Score of 24.85 and 31 stars. This unofficial implementation of DeepSeek Engram enables the injection of high-capacity conditional memory into LLMs via sparse retrieval PEFT, without increasing inference FLOPs. Its growing popularity stems from the potential for improving LLM performance while maintaining efficiency.
ZJU-OmniAI/GFT has a Growth Score of 16.72 and 28 stars, and offers a novel approach to fine-tuning called GFT, which leverages unbiased group advantages and dynamic coefficient rectification. This repository is gaining traction due to its potential for improving the stability and effectiveness of reward-based fine-tuning methods.
UNfukashigi/Anima-LoRA-Factory has a Growth Score of 14.50 and 21 stars, providing a user-friendly GUI tool for training LoRAs (Low-Rank Adaptation) for next-generation Anima diffusion models. Its growth can be attributed to the increasing popularity of LoRA as a fine-tuning technique and the need for accessible tools.
WillowHe/EvoOpt_oppangu_optimization_model has a Growth Score of 10.29 and an impressive 514 stars, offering solutions for leveraging Openpangu-7B as a base model for fine-tuning and applying large language models to operations research optimization tasks. Its popularity stems from the growing interest in using LLMs for real-world applications.
SUM-INNOVATION/RUMUS has a Growth Score of 9.80 and 163 stars, providing a Rust-based framework for training neural networks. Its growth can be attributed to the increasing adoption of Rust as a programming language for AI development.
semidark/kokoro-deutsch has a Growth Score of 7.87 and 29 stars, offering a complete recipe for fine-tuning Kokoro-82M on new languages, in this case German. This repository is gaining traction due to its potential for improving multilingual support in LLMs.
Other notable mentions include Dynamis-Labs/spectralquant, which explores spectral structure to improve compression limits; Mintzs/oogaboogalm, which proposes fine-tuning AI models with caveman system prompts; and PentesterFlow/OffensiveSET, which generates high-quality pentesting conversation datasets for LLM fine-tuning.