Today's Fine-tuning & Training: Fastest-Growing Projects — April 25, 2026
Today's the Fine-tuning & Training space, we're seeing a surge of interest in tools that enable efficient and effective fine-tuning of large language models (LLMs) for various tasks. With many repositories focusing on improving the performance of LLMs with techniques like sparse retrieval and unbiased group advantages, it's clear that researchers are eager to push the boundaries of what these models can achieve.
One standout repository is mattmireles/gemma-tuner-multimodal, which has seen a significant growth score of 63.31 and now boasts 1,385 stars. This tool enables fine-tuning of Gemma 4 and 3n with audio, images, and text on Apple Silicon using PyTorch and Metal Performance Shaders, making it an attractive option for those looking to explore multimodal learning. Its high growth score suggests that the community is eager to leverage its capabilities for a wide range of applications.
Another notable repository is QingGo/engram-peft, which has achieved a growth score of 24.85 and garnered 31 stars. This unofficial implementation of DeepSeek Engram allows users to inject high-capacity conditional memory into LLMs via sparse retrieval PEFT without increasing inference FLOPs. Its growing popularity indicates that researchers are interested in exploring new ways to enhance the capabilities of LLMs.
ZJU-OmniAI/GFT is another repository worth mentioning, with a growth score of 16.67 and 27 stars. This tool focuses on fine-tuning with unbiased group advantages and dynamic coefficient rectification, offering a novel approach to improving the performance of LLMs. Its growth suggests that researchers are looking for new techniques to address the challenges of fine-tuning.
WillowHe/EvoOpt_oppangu_optimization_model has seen significant interest, with 514 stars and a growth score of 10.29. This repository provides solutions leveraging Openpangu - 7B as the base model for fine-tuning and application of LLMs in operations research optimization tasks. Its popularity indicates that researchers are exploring new ways to apply LLMs to real-world problems.
Other notable repositories include SUM-INNOVATION/RUMUS, a Rust-based framework to train neural networks with a growth score of 9.66 and 156 stars; semidark/kokoro-deutsch, which provides a complete training recipe for fine-tuning Kokoro-82M on German with a growth score of 7.87 and 29 stars; and Dynamis-Labs/spectralquant, which breaks TurboQuant's compression limit via spectral structure with a growth score of 4.72 and 126 stars.
PentesterFlow/OffensiveSET is also worth mentioning, with a growth score of 3.50 and 71 stars. This repository generates high-quality pentesting conversation datasets for LLM fine-tuning, highlighting the growing interest in applying LLMs to cybersecurity tasks.
One standout repository is mattmireles/gemma-tuner-multimodal, which has seen a significant growth score of 63.31 and now boasts 1,385 stars. This tool enables fine-tuning of Gemma 4 and 3n with audio, images, and text on Apple Silicon using PyTorch and Metal Performance Shaders, making it an attractive option for those looking to explore multimodal learning. Its high growth score suggests that the community is eager to leverage its capabilities for a wide range of applications.
Another notable repository is QingGo/engram-peft, which has achieved a growth score of 24.85 and garnered 31 stars. This unofficial implementation of DeepSeek Engram allows users to inject high-capacity conditional memory into LLMs via sparse retrieval PEFT without increasing inference FLOPs. Its growing popularity indicates that researchers are interested in exploring new ways to enhance the capabilities of LLMs.
ZJU-OmniAI/GFT is another repository worth mentioning, with a growth score of 16.67 and 27 stars. This tool focuses on fine-tuning with unbiased group advantages and dynamic coefficient rectification, offering a novel approach to improving the performance of LLMs. Its growth suggests that researchers are looking for new techniques to address the challenges of fine-tuning.
WillowHe/EvoOpt_oppangu_optimization_model has seen significant interest, with 514 stars and a growth score of 10.29. This repository provides solutions leveraging Openpangu - 7B as the base model for fine-tuning and application of LLMs in operations research optimization tasks. Its popularity indicates that researchers are exploring new ways to apply LLMs to real-world problems.
Other notable repositories include SUM-INNOVATION/RUMUS, a Rust-based framework to train neural networks with a growth score of 9.66 and 156 stars; semidark/kokoro-deutsch, which provides a complete training recipe for fine-tuning Kokoro-82M on German with a growth score of 7.87 and 29 stars; and Dynamis-Labs/spectralquant, which breaks TurboQuant's compression limit via spectral structure with a growth score of 4.72 and 126 stars.
PentesterFlow/OffensiveSET is also worth mentioning, with a growth score of 3.50 and 71 stars. This repository generates high-quality pentesting conversation datasets for LLM fine-tuning, highlighting the growing interest in applying LLMs to cybersecurity tasks.