Today's Fine-tuning & Training: Fastest-Growing Projects — April 29, 2026
Today's the Fine-tuning & Training space, we're seeing a surge in interest around multimodal fine-tuning and tools that enable users to train models on specific tasks or domains. The top-growing repositories are showcasing innovative approaches to fine-tuning large language models (LLMs) and other AI architectures, often leveraging PyTorch, Metal Performance Shaders, or other optimized frameworks.
mattmireles/gemma-tuner-multimodal is a standout repository with a growth score of 52.23 and over 1,395 stars. This tool allows users to fine-tune Gemma 4 and 3n models on Apple Silicon using PyTorch and Metal Performance Shaders, enabling efficient multimodal training with audio, images, and text inputs. Its popularity is likely due to the growing demand for multimodal AI applications.
QingGo/engram-peft has a growth score of 19.00 and 31 stars. This repository provides an unofficial implementation of DeepSeek Engram, which injects high-capacity conditional memory into LLMs via sparse retrieval PEFT without increasing inference FLOPs. Its growing interest can be attributed to the potential for improving LLM performance in specific tasks.
UNfukashigi/Anima-LoRA-Factory boasts a growth score of 14.20 and 23 stars, offering a user-friendly GUI tool for training LoRAs for next-generation Anima diffusion models. The repository's popularity likely stems from its ease of use and the growing interest in diffusion-based AI architectures.
ZJU-OmniAI/GFT has a growth score of 11.73 and 29 stars, presenting a framework called GFT that enables fine-tuning with unbiased group advantages and dynamic coefficient rectification. Its growing attention may be due to its innovative approach to addressing common challenges in imitation learning and reward fine-tuning.
SUM-INNOVATION/RUMUS features a growth score of 9.47 and an impressive 201 stars, providing a Rust-based framework for training neural networks. The repository's popularity likely stems from the increasing demand for efficient and scalable AI development tools.
semidark/kokoro-deutsch has a growth score of 6.39 and 30 stars, offering a complete training recipe for fine-tuning Kokoro-82M on German language data. Its growing interest may be attributed to the need for more diverse language models.
Dynamis-Labs/spectralquant boasts a growth score of 4.06 and 129 stars, showcasing research on breaking TurboQuant's compression limit via spectral structure. Although its commit activity has been low, its popularity likely stems from its innovative approach to neural network quantization.
Mintzs/oogaboogalm features a growth score of 3.16 and 45 stars, exploring the idea of fine-tuning AI models with baked-in skills and prompts for reduced token use. The repository's growing attention may be due to its potential applications in resource-constrained environments.
PentesterFlow/OffensiveSET has a growth score of 3.02 and 73 stars, providing an MCP server for generating high-quality pentesting conversation datasets for LLM fine-tuning. Its popularity likely stems from the increasing demand for robust security testing tools.
Goekdeniz-Guelmez/moshi-finetune-mlx rounds out our list with a growth score of 0.71 and 25 stars, offering a tool for fine-tuning Moshi speech-to-speech models on Apple Silicon. Although its growth has been relatively slow, it may attract attention from developers interested in real-time speech processing applications.
Overall, Today's Fine-tuning & Training landscape is marked by innovative approaches to multimodal training, LLM fine-tuning, and AI development frameworks.
mattmireles/gemma-tuner-multimodal is a standout repository with a growth score of 52.23 and over 1,395 stars. This tool allows users to fine-tune Gemma 4 and 3n models on Apple Silicon using PyTorch and Metal Performance Shaders, enabling efficient multimodal training with audio, images, and text inputs. Its popularity is likely due to the growing demand for multimodal AI applications.
QingGo/engram-peft has a growth score of 19.00 and 31 stars. This repository provides an unofficial implementation of DeepSeek Engram, which injects high-capacity conditional memory into LLMs via sparse retrieval PEFT without increasing inference FLOPs. Its growing interest can be attributed to the potential for improving LLM performance in specific tasks.
UNfukashigi/Anima-LoRA-Factory boasts a growth score of 14.20 and 23 stars, offering a user-friendly GUI tool for training LoRAs for next-generation Anima diffusion models. The repository's popularity likely stems from its ease of use and the growing interest in diffusion-based AI architectures.
ZJU-OmniAI/GFT has a growth score of 11.73 and 29 stars, presenting a framework called GFT that enables fine-tuning with unbiased group advantages and dynamic coefficient rectification. Its growing attention may be due to its innovative approach to addressing common challenges in imitation learning and reward fine-tuning.
SUM-INNOVATION/RUMUS features a growth score of 9.47 and an impressive 201 stars, providing a Rust-based framework for training neural networks. The repository's popularity likely stems from the increasing demand for efficient and scalable AI development tools.
semidark/kokoro-deutsch has a growth score of 6.39 and 30 stars, offering a complete training recipe for fine-tuning Kokoro-82M on German language data. Its growing interest may be attributed to the need for more diverse language models.
Dynamis-Labs/spectralquant boasts a growth score of 4.06 and 129 stars, showcasing research on breaking TurboQuant's compression limit via spectral structure. Although its commit activity has been low, its popularity likely stems from its innovative approach to neural network quantization.
Mintzs/oogaboogalm features a growth score of 3.16 and 45 stars, exploring the idea of fine-tuning AI models with baked-in skills and prompts for reduced token use. The repository's growing attention may be due to its potential applications in resource-constrained environments.
PentesterFlow/OffensiveSET has a growth score of 3.02 and 73 stars, providing an MCP server for generating high-quality pentesting conversation datasets for LLM fine-tuning. Its popularity likely stems from the increasing demand for robust security testing tools.
Goekdeniz-Guelmez/moshi-finetune-mlx rounds out our list with a growth score of 0.71 and 25 stars, offering a tool for fine-tuning Moshi speech-to-speech models on Apple Silicon. Although its growth has been relatively slow, it may attract attention from developers interested in real-time speech processing applications.
Overall, Today's Fine-tuning & Training landscape is marked by innovative approaches to multimodal training, LLM fine-tuning, and AI development frameworks.