Today's Fine-tuning & Training: Fastest-Growing Projects — April 27, 2026
Today's Fine-tuning & Training, we're seeing a surge in innovative approaches to optimizing and fine-tuning large language models (LLMs) and neural networks. Multimodal learning, sparse retrieval, and conditional memory injection are just a few of the trends that dominated the space this week. With many repositories showcasing impressive growth scores, it's clear that developers are eager to explore new ways to improve model performance.
mattmireles/gemma-tuner-multimodal takes the top spot with an impressive Growth Score of 57.23 and over 1,389 stars. This repository provides a framework for fine-tuning Gemma models on Apple Silicon using PyTorch and Metal Performance Shaders, allowing for efficient processing of audio, images, and text inputs. Its growth can be attributed to the increasing demand for multimodal learning capabilities in AI applications.
QingGo/engram-peft boasts a Growth Score of 21.53 and 31 stars, showcasing an unofficial implementation of DeepSeek Engram that injects high-capacity conditional memory into LLMs via sparse retrieval PEFT. This innovative approach to model optimization has captured the attention of developers looking for ways to enhance their models' performance without increasing inference FLOPs.
UNfukashigi/Anima-LoRA-Factory, with a Growth Score of 17.75 and 23 stars, offers a user-friendly GUI tool for training LoRAs for next-generation Anima diffusion models. The growth of this repository highlights the growing interest in LoRA (Low-Rank Adaptation) as an efficient fine-tuning technique for large language models.
ZJU-OmniAI/GFT has achieved a Growth Score of 13.73 and 29 stars by providing a framework for reward fine-tuning with unbiased group advantages and dynamic coefficient rectification, showcasing the ongoing efforts to improve reinforcement learning methods. Its growth indicates the increasing demand for more efficient and effective training techniques.
SUM-INNOVATION/RUMUS boasts a Growth Score of 9.65 and 185 stars as a Rust-based framework for training neural networks, demonstrating the growing interest in using Rust for machine learning applications. The growth of this repository highlights the need for efficient and scalable frameworks that can handle complex neural network architectures.
WillowHe/EvoOpt_oppangu_optimization_model has gained significant attention with a Growth Score of 9.55 and 514 stars by leveraging Openpangu-7B as a base model for fine-tuning and applying LLMs to operations research optimization tasks. This growth reflects the increasing interest in using large language models for complex optimization problems.
semidark/kokoro-deutsch showcases a complete training recipe for fine-tuning Kokoro-82M on German, with a Growth Score of 7.03 and 29 stars. The growth of this repository highlights the ongoing efforts to adapt pre-trained language models to new languages and domains.
Dynamis-Labs/spectralquant has achieved a Growth Score of 4.39 and 127 stars by introducing a novel method for breaking compression limits via spectral structure, demonstrating innovative approaches to model optimization. Its growth indicates the increasing interest in exploring new techniques for improving model performance.
Mintzs/oogaboogalm proposes an intriguing approach to fine-tuning AI models using caveman system prompts and skills, with a Growth Score of 3.53 and 45 stars. Although not as widely adopted as other repositories on this list, its growth reflects the ongoing efforts to reduce token usage in LLMs.
PentesterFlow/OffensiveSET rounds out our list with a Growth Score of 3.22 and 71 stars, providing a dataset generator for high-quality pentesting conversation datasets used in LLM fine-tuning. Its growth highlights the increasing importance of high-quality training data in improving model performance.
mattmireles/gemma-tuner-multimodal takes the top spot with an impressive Growth Score of 57.23 and over 1,389 stars. This repository provides a framework for fine-tuning Gemma models on Apple Silicon using PyTorch and Metal Performance Shaders, allowing for efficient processing of audio, images, and text inputs. Its growth can be attributed to the increasing demand for multimodal learning capabilities in AI applications.
QingGo/engram-peft boasts a Growth Score of 21.53 and 31 stars, showcasing an unofficial implementation of DeepSeek Engram that injects high-capacity conditional memory into LLMs via sparse retrieval PEFT. This innovative approach to model optimization has captured the attention of developers looking for ways to enhance their models' performance without increasing inference FLOPs.
UNfukashigi/Anima-LoRA-Factory, with a Growth Score of 17.75 and 23 stars, offers a user-friendly GUI tool for training LoRAs for next-generation Anima diffusion models. The growth of this repository highlights the growing interest in LoRA (Low-Rank Adaptation) as an efficient fine-tuning technique for large language models.
ZJU-OmniAI/GFT has achieved a Growth Score of 13.73 and 29 stars by providing a framework for reward fine-tuning with unbiased group advantages and dynamic coefficient rectification, showcasing the ongoing efforts to improve reinforcement learning methods. Its growth indicates the increasing demand for more efficient and effective training techniques.
SUM-INNOVATION/RUMUS boasts a Growth Score of 9.65 and 185 stars as a Rust-based framework for training neural networks, demonstrating the growing interest in using Rust for machine learning applications. The growth of this repository highlights the need for efficient and scalable frameworks that can handle complex neural network architectures.
WillowHe/EvoOpt_oppangu_optimization_model has gained significant attention with a Growth Score of 9.55 and 514 stars by leveraging Openpangu-7B as a base model for fine-tuning and applying LLMs to operations research optimization tasks. This growth reflects the increasing interest in using large language models for complex optimization problems.
semidark/kokoro-deutsch showcases a complete training recipe for fine-tuning Kokoro-82M on German, with a Growth Score of 7.03 and 29 stars. The growth of this repository highlights the ongoing efforts to adapt pre-trained language models to new languages and domains.
Dynamis-Labs/spectralquant has achieved a Growth Score of 4.39 and 127 stars by introducing a novel method for breaking compression limits via spectral structure, demonstrating innovative approaches to model optimization. Its growth indicates the increasing interest in exploring new techniques for improving model performance.
Mintzs/oogaboogalm proposes an intriguing approach to fine-tuning AI models using caveman system prompts and skills, with a Growth Score of 3.53 and 45 stars. Although not as widely adopted as other repositories on this list, its growth reflects the ongoing efforts to reduce token usage in LLMs.
PentesterFlow/OffensiveSET rounds out our list with a Growth Score of 3.22 and 71 stars, providing a dataset generator for high-quality pentesting conversation datasets used in LLM fine-tuning. Its growth highlights the increasing importance of high-quality training data in improving model performance.