Today's Fine-tuning & Training: Fastest-Growing Projects — April 18, 2026
Today's Fine-tuning & Training space saw significant activity around optimizing large language models (LLMs) for better performance and efficiency. The trend is clear: researchers are focusing on fine-tuning and training techniques that can unlock faster, more accurate, and more memory-efficient LLMs.
Facebookresearch/tribev2 takes the top spot with a Growth Score of 63.52 and 1,892 stars. This repository contains code to train and evaluate TRIBE v2, a multimodal model for brain response prediction, demonstrating growing interest in using AI for neuroscience applications. Its high growth score indicates that researchers are eager to explore this innovative approach.
QingGo/engram-peft comes in second with a Growth Score of 34.33 and 25 stars. This unofficial implementation of DeepSeek Engram injects high-capacity conditional memory into LLMs via sparse retrieval PEFT without increasing inference FLOPs, making it an attractive solution for those looking to boost their models' performance. Its rapid growth suggests that researchers are eager to explore new ways to optimize LLMs.
0xSero/turboquant boasts a Growth Score of 30.69 and 1,086 stars. This project offers near-optimal KV cache quantization for LLM inference with Triton kernels + vLLM integration, showcasing the growing interest in optimizing LLM performance. Its high growth score indicates that researchers are eager to adopt this efficient solution.
tonbistudio/turboquant-pytorch has a Growth Score of 28.27 and 943 stars. This from-scratch PyTorch implementation of Google's TurboQuant for LLM KV cache compression achieves impressive results, offering 5x compression at 3-bit with 99.5% attention fidelity. Its growth suggests that researchers are looking for reliable and efficient solutions for their LLMs.
WillowHe/EvoOpt_oppangu_optimization_model has a Growth Score of 9.45 and 338 stars. This project provides a set of solutions leveraging Openpangu - 7B as the base model for fine-tuning and application of large language models (LLMs) in operations research optimization tasks, indicating growing interest in using AI for specific industries. Its growth score indicates moderate interest from researchers.
SUM-INNOVATION/RUMUS has a Growth Score of 8.86 and 70 stars. This Rust-based framework to train Neural Networks offers an alternative solution for those looking for more efficient training methods. Its growth suggests that researchers are exploring different programming languages for their AI projects.
Mintzs/oogaboogalm boasts a Growth Score of 7.19 and 40 stars. This project explores fine-tuning models with caveman system prompts and skills to reduce token use, showcasing innovative approaches to optimize LLMs. Its growth score indicates moderate interest from researchers.
OnlyTerp/turboquant has a Growth Score of 6.92 and 53 stars. This open-source implementation of Google TurboQuant for near-optimal KV cache compression offers an efficient solution for LLM inference. Its growth suggests that researchers are looking for reliable and efficient solutions for their LLMs.
Dynamis-Labs/spectralquant rounds out the list with a Growth Score of 6.58 and 111 stars. This project breaks TurboQuant's compression limit via spectral structure, showcasing innovative approaches to optimize LLM performance. Its growth score indicates moderate interest from researchers.
Facebookresearch/tribev2 takes the top spot with a Growth Score of 63.52 and 1,892 stars. This repository contains code to train and evaluate TRIBE v2, a multimodal model for brain response prediction, demonstrating growing interest in using AI for neuroscience applications. Its high growth score indicates that researchers are eager to explore this innovative approach.
QingGo/engram-peft comes in second with a Growth Score of 34.33 and 25 stars. This unofficial implementation of DeepSeek Engram injects high-capacity conditional memory into LLMs via sparse retrieval PEFT without increasing inference FLOPs, making it an attractive solution for those looking to boost their models' performance. Its rapid growth suggests that researchers are eager to explore new ways to optimize LLMs.
0xSero/turboquant boasts a Growth Score of 30.69 and 1,086 stars. This project offers near-optimal KV cache quantization for LLM inference with Triton kernels + vLLM integration, showcasing the growing interest in optimizing LLM performance. Its high growth score indicates that researchers are eager to adopt this efficient solution.
tonbistudio/turboquant-pytorch has a Growth Score of 28.27 and 943 stars. This from-scratch PyTorch implementation of Google's TurboQuant for LLM KV cache compression achieves impressive results, offering 5x compression at 3-bit with 99.5% attention fidelity. Its growth suggests that researchers are looking for reliable and efficient solutions for their LLMs.
WillowHe/EvoOpt_oppangu_optimization_model has a Growth Score of 9.45 and 338 stars. This project provides a set of solutions leveraging Openpangu - 7B as the base model for fine-tuning and application of large language models (LLMs) in operations research optimization tasks, indicating growing interest in using AI for specific industries. Its growth score indicates moderate interest from researchers.
SUM-INNOVATION/RUMUS has a Growth Score of 8.86 and 70 stars. This Rust-based framework to train Neural Networks offers an alternative solution for those looking for more efficient training methods. Its growth suggests that researchers are exploring different programming languages for their AI projects.
Mintzs/oogaboogalm boasts a Growth Score of 7.19 and 40 stars. This project explores fine-tuning models with caveman system prompts and skills to reduce token use, showcasing innovative approaches to optimize LLMs. Its growth score indicates moderate interest from researchers.
OnlyTerp/turboquant has a Growth Score of 6.92 and 53 stars. This open-source implementation of Google TurboQuant for near-optimal KV cache compression offers an efficient solution for LLM inference. Its growth suggests that researchers are looking for reliable and efficient solutions for their LLMs.
Dynamis-Labs/spectralquant rounds out the list with a Growth Score of 6.58 and 111 stars. This project breaks TurboQuant's compression limit via spectral structure, showcasing innovative approaches to optimize LLM performance. Its growth score indicates moderate interest from researchers.