Today's Fine-tuning & Training: Fastest-Growing Projects — May 08, 2026
This week, we're seeing significant activity in the Fine-tuning & Training space on GitHub, with a focus on innovative methods for injecting high-capacity conditional memory into large language models (LLMs) and user-friendly tools for training LoRAs. Additionally, there's growing interest in fine-tuning frameworks for specific tasks, such as medical binary classification.
QingGo/engram-peft is making waves with its unofficial implementation of DeepSeek Engram, which enables the injection of high-capacity conditional memory into LLMs via sparse retrieval PEFT without increasing inference FLOPs. With a growth score of 12.42 and 31 stars, this repository is gaining traction due to its potential to revolutionize the way we fine-tune LLMs.
UNfukashigi/Anima-LoRA-Factory offers a user-friendly GUI tool for training LoRAs for next-generation Anima diffusion models, making it easier for developers to work with these complex models. Its growth score of 7.89 and 27 stars reflect the growing interest in LoRAs and the need for more accessible tools.
ZJU-OmniAI/GFT presents a novel approach to fine-tuning, focusing on unbiased group advantages and dynamic coefficient rectification. With a growth score of 6.95 and 30 stars, this repository is attracting attention from researchers looking for new methods to improve their fine-tuning results.
Semidark/kokoro-deutsch provides a complete training recipe for fine-tuning Kokoro-82M on German, demonstrating the growing interest in multilingual models. Its growth score of 5.09 and 33 stars reflect the demand for more language-specific fine-tuning recipes.
Raiyanyahya/how-to-train-your-gpt is an extremely popular repository with a growth score of 5.03 and an impressive 730 stars, offering a commented and explained guide to building a modern LLM from scratch. Its enduring popularity stems from its accessibility and the growing interest in building custom LLMs.
Jackohhhh/MedLLM-Finetuning offers an out-of-the-box fine-tuning framework for medical binary classification tasks, catering to the increasing demand for specialized fine-tuning solutions. With a growth score of 3.34 and 21 stars, this repository is gaining traction among developers working on medical applications.
Mintzs/oogaboogalm explores the idea of baking caveman system prompts into LLMs through fine-tuning, sparking interest in novel approaches to model optimization. Its growth score of 2.18 and 47 stars reflect the curiosity surrounding this innovative concept.
VVVVVjdy/D-OPSD presents an official repository for on-policy self-distillation for continuously tuning step-distilled diffusion models, showcasing the ongoing research into more efficient fine-tuning methods. With a growth score of 1.41 and 52 stars, this repository is attracting attention from researchers in the field.
Hlpun/Train-in-Silence offers a task-aware MCP server and automated VRAM calculator for LLM fine-tuning, addressing the growing need for optimized training workflows. Its growth score of 1.33 and 39 stars reflect the interest in streamlining fine-tuning processes.
QingGo/engram-peft is making waves with its unofficial implementation of DeepSeek Engram, which enables the injection of high-capacity conditional memory into LLMs via sparse retrieval PEFT without increasing inference FLOPs. With a growth score of 12.42 and 31 stars, this repository is gaining traction due to its potential to revolutionize the way we fine-tune LLMs.
UNfukashigi/Anima-LoRA-Factory offers a user-friendly GUI tool for training LoRAs for next-generation Anima diffusion models, making it easier for developers to work with these complex models. Its growth score of 7.89 and 27 stars reflect the growing interest in LoRAs and the need for more accessible tools.
ZJU-OmniAI/GFT presents a novel approach to fine-tuning, focusing on unbiased group advantages and dynamic coefficient rectification. With a growth score of 6.95 and 30 stars, this repository is attracting attention from researchers looking for new methods to improve their fine-tuning results.
Semidark/kokoro-deutsch provides a complete training recipe for fine-tuning Kokoro-82M on German, demonstrating the growing interest in multilingual models. Its growth score of 5.09 and 33 stars reflect the demand for more language-specific fine-tuning recipes.
Raiyanyahya/how-to-train-your-gpt is an extremely popular repository with a growth score of 5.03 and an impressive 730 stars, offering a commented and explained guide to building a modern LLM from scratch. Its enduring popularity stems from its accessibility and the growing interest in building custom LLMs.
Jackohhhh/MedLLM-Finetuning offers an out-of-the-box fine-tuning framework for medical binary classification tasks, catering to the increasing demand for specialized fine-tuning solutions. With a growth score of 3.34 and 21 stars, this repository is gaining traction among developers working on medical applications.
Mintzs/oogaboogalm explores the idea of baking caveman system prompts into LLMs through fine-tuning, sparking interest in novel approaches to model optimization. Its growth score of 2.18 and 47 stars reflect the curiosity surrounding this innovative concept.
VVVVVjdy/D-OPSD presents an official repository for on-policy self-distillation for continuously tuning step-distilled diffusion models, showcasing the ongoing research into more efficient fine-tuning methods. With a growth score of 1.41 and 52 stars, this repository is attracting attention from researchers in the field.
Hlpun/Train-in-Silence offers a task-aware MCP server and automated VRAM calculator for LLM fine-tuning, addressing the growing need for optimized training workflows. Its growth score of 1.33 and 39 stars reflect the interest in streamlining fine-tuning processes.