Today's Fine-tuning & Training: Fastest-Growing Projects — April 11, 2026
Today's the Fine-tuning & Training space, we saw a surge of interest in tools related to multimodal models and large language model (LLM) optimization. Many of these repositories focus on improving the efficiency and performance of LLMs, with several implementations of Google's TurboQuant algorithm making an appearance.
Facebookresearch/tribev2 is gaining traction with a growth score of 82.03 and 1,750 stars, as it provides code to train and evaluate TRIBE v2, a multimodal model for brain response prediction. This repository is likely growing due to the increasing interest in multimodal models and their applications in various fields.
Tonbistudio/turboquant-pytorch has a growth score of 37.97 and 895 stars, offering a from-scratch PyTorch implementation of Google's TurboQuant for LLM KV cache compression. Its popularity can be attributed to the need for efficient compression methods in large language models, with this implementation achieving 5x compression at 3-bit with 99.5% attention fidelity.
0xSero/turboquant boasts a growth score of 37.91 and 941 stars, featuring TurboQuant: near-optimal KV cache quantization for LLM inference with Triton kernels + vLLM integration. This repository's growth is likely driven by the demand for optimized LLM inference methods, particularly those that leverage Triton kernels.
Dynamis-Labs/spectralquant has a growth score of 13.17 and 101 stars, presenting a method to break TurboQuant's compression limit via spectral structure. Its growing popularity can be attributed to the ongoing research in optimizing LLM compression methods, with this approach promising improved results.
TYH-labs/unsloth-buddy boasts a growth score of 12.11 and 216 stars, offering zero-friction LLM fine-tuning skill for Claude Code, Gemini CLI & any ACP agent. This repository's growth is likely driven by the need for streamlined fine-tuning processes in various AI platforms.
WillowHe/EvoOpt_oppangu_optimization_model has a growth score of 11.46 and 254 stars, providing solutions for fine-tuning large language models (LLMs) in operations research optimization tasks using Openpangu - 7B as the base model. Its growing popularity can be attributed to the increasing interest in applying LLMs to operations research.
PentesterFlow/OffensiveSET features a growth score of 7.67 and 60 stars, offering an Offensive Security Dataset Generator for generating high-quality pentesting conversation datasets for LLM fine-tuning. This repository's growth is likely driven by the need for diverse and realistic datasets in AI security applications.
OnlyTerp/turboquant boasts a growth score of 6.91 and 52 stars, presenting the first open-source implementation of Google TurboQuant -- near-optimal KV cache compression for LLM inference. Its popularity can be attributed to the ongoing research in optimizing LLM compression methods, with this approach promising near-zero quality loss.
Lastly, mattmireles/gemma-tuner-multimodal has a growth score of 4.26 and an undisclosed number of stars, focusing on fine-tuning Gemma 4 and 3n with audio, images, and text on Apple Silicon using PyTorch and Metal Performance Shaders. Although its description is brief, this repository's growth may be driven by the increasing interest in multimodal models and their applications on specific hardware platforms.
Facebookresearch/tribev2 is gaining traction with a growth score of 82.03 and 1,750 stars, as it provides code to train and evaluate TRIBE v2, a multimodal model for brain response prediction. This repository is likely growing due to the increasing interest in multimodal models and their applications in various fields.
Tonbistudio/turboquant-pytorch has a growth score of 37.97 and 895 stars, offering a from-scratch PyTorch implementation of Google's TurboQuant for LLM KV cache compression. Its popularity can be attributed to the need for efficient compression methods in large language models, with this implementation achieving 5x compression at 3-bit with 99.5% attention fidelity.
0xSero/turboquant boasts a growth score of 37.91 and 941 stars, featuring TurboQuant: near-optimal KV cache quantization for LLM inference with Triton kernels + vLLM integration. This repository's growth is likely driven by the demand for optimized LLM inference methods, particularly those that leverage Triton kernels.
Dynamis-Labs/spectralquant has a growth score of 13.17 and 101 stars, presenting a method to break TurboQuant's compression limit via spectral structure. Its growing popularity can be attributed to the ongoing research in optimizing LLM compression methods, with this approach promising improved results.
TYH-labs/unsloth-buddy boasts a growth score of 12.11 and 216 stars, offering zero-friction LLM fine-tuning skill for Claude Code, Gemini CLI & any ACP agent. This repository's growth is likely driven by the need for streamlined fine-tuning processes in various AI platforms.
WillowHe/EvoOpt_oppangu_optimization_model has a growth score of 11.46 and 254 stars, providing solutions for fine-tuning large language models (LLMs) in operations research optimization tasks using Openpangu - 7B as the base model. Its growing popularity can be attributed to the increasing interest in applying LLMs to operations research.
PentesterFlow/OffensiveSET features a growth score of 7.67 and 60 stars, offering an Offensive Security Dataset Generator for generating high-quality pentesting conversation datasets for LLM fine-tuning. This repository's growth is likely driven by the need for diverse and realistic datasets in AI security applications.
OnlyTerp/turboquant boasts a growth score of 6.91 and 52 stars, presenting the first open-source implementation of Google TurboQuant -- near-optimal KV cache compression for LLM inference. Its popularity can be attributed to the ongoing research in optimizing LLM compression methods, with this approach promising near-zero quality loss.
Lastly, mattmireles/gemma-tuner-multimodal has a growth score of 4.26 and an undisclosed number of stars, focusing on fine-tuning Gemma 4 and 3n with audio, images, and text on Apple Silicon using PyTorch and Metal Performance Shaders. Although its description is brief, this repository's growth may be driven by the increasing interest in multimodal models and their applications on specific hardware platforms.