Today's Fine-tuning & Training: Fastest-Growing Projects — April 30, 2026
Today's the Fine-tuning & Training space, we're seeing a surge in innovative approaches to optimizing language models and leveraging multimodal inputs. The trend is shifting towards more efficient and effective fine-tuning methods, with many repositories showcasing novel techniques for injecting high-capacity conditional memory into LLMs or exploring new architectures for diffusion models.
mattmireles/gemma-tuner-multimodal takes the top spot with a Growth Score of 50.00 and 1,397 stars, as it enables fine-tuning of Gemma 4 and 3n models on Apple Silicon using PyTorch and Metal Performance Shaders, making it an attractive solution for those looking to optimize their models for multimodal inputs.
QingGo/engram-peft boasts a Growth Score of 17.94 and 31 stars, with its unofficial implementation of DeepSeek Engram allowing users to inject high-capacity conditional memory into LLMs via sparse retrieval PEFT without increasing inference FLOPs, making it an appealing option for those seeking to enhance their models' capabilities.
UNfukashigi/Anima-LoRA-Factory has a Growth Score of 12.91 and 23 stars, offering a user-friendly GUI tool designed specifically for training LoRAs for next-generation Anima diffusion models, which is likely driving its growth as more developers explore this emerging area.
ZJU-OmniAI/GFT has a Growth Score of 10.89 and 29 stars, presenting a novel approach to fine-tuning with unbiased group advantages and dynamic coefficient rectification, which may be attracting attention from researchers looking for innovative methods to improve their models' performance.
semidark/kokoro-deutsch boasts a Growth Score of 6.17 and 31 stars, providing a complete training recipe for fine-tuning Kokoro-82M on German language data, making it a valuable resource for those working with this specific model and language combination.
Dynamis-Labs/spectralquant has a Growth Score of 3.90 and 129 stars, breaking new ground in the field of quantization by proposing a spectral structure-based approach that achieves state-of-the-art results with minimal overhead, which may be driving its growth despite relatively few recent commits.
Mintzs/oogaboogalm has a Growth Score of 3.02 and 46 stars, exploring an intriguing idea of baking caveman system prompts and skills into AI models through fine-tuning, which could be attracting interest from developers seeking novel approaches to model optimization.
PentesterFlow/OffensiveSET has a Growth Score of 2.91 and 73 stars, offering a unique dataset generator for generating high-quality pentesting conversation datasets for LLM fine-tuning, making it a valuable resource for those working in the field of cybersecurity.
Goekdeniz-Guelmez/moshi-finetune-mlx rounds out our list with a Growth Score of 0.69 and 25 stars, providing a fine-tuning solution for Moshi models on Apple Silicon, which may be attracting attention from developers looking to optimize their speech-to-speech models.
mattmireles/gemma-tuner-multimodal takes the top spot with a Growth Score of 50.00 and 1,397 stars, as it enables fine-tuning of Gemma 4 and 3n models on Apple Silicon using PyTorch and Metal Performance Shaders, making it an attractive solution for those looking to optimize their models for multimodal inputs.
QingGo/engram-peft boasts a Growth Score of 17.94 and 31 stars, with its unofficial implementation of DeepSeek Engram allowing users to inject high-capacity conditional memory into LLMs via sparse retrieval PEFT without increasing inference FLOPs, making it an appealing option for those seeking to enhance their models' capabilities.
UNfukashigi/Anima-LoRA-Factory has a Growth Score of 12.91 and 23 stars, offering a user-friendly GUI tool designed specifically for training LoRAs for next-generation Anima diffusion models, which is likely driving its growth as more developers explore this emerging area.
ZJU-OmniAI/GFT has a Growth Score of 10.89 and 29 stars, presenting a novel approach to fine-tuning with unbiased group advantages and dynamic coefficient rectification, which may be attracting attention from researchers looking for innovative methods to improve their models' performance.
semidark/kokoro-deutsch boasts a Growth Score of 6.17 and 31 stars, providing a complete training recipe for fine-tuning Kokoro-82M on German language data, making it a valuable resource for those working with this specific model and language combination.
Dynamis-Labs/spectralquant has a Growth Score of 3.90 and 129 stars, breaking new ground in the field of quantization by proposing a spectral structure-based approach that achieves state-of-the-art results with minimal overhead, which may be driving its growth despite relatively few recent commits.
Mintzs/oogaboogalm has a Growth Score of 3.02 and 46 stars, exploring an intriguing idea of baking caveman system prompts and skills into AI models through fine-tuning, which could be attracting interest from developers seeking novel approaches to model optimization.
PentesterFlow/OffensiveSET has a Growth Score of 2.91 and 73 stars, offering a unique dataset generator for generating high-quality pentesting conversation datasets for LLM fine-tuning, making it a valuable resource for those working in the field of cybersecurity.
Goekdeniz-Guelmez/moshi-finetune-mlx rounds out our list with a Growth Score of 0.69 and 25 stars, providing a fine-tuning solution for Moshi models on Apple Silicon, which may be attracting attention from developers looking to optimize their speech-to-speech models.