Zenyx_114M-Tiny-Edu-Instruct is an experimental small instruction fine-tuned language model with approximately 114 million parameters. It is built on the TinyEdu-50M base model, pre-trained on the FineWeb-Edu dataset, and fine-tuned on a mixed dataset of OpenHermes-2.5 and CodeFeedback-Filtered. This model aims to explore the limits of instruction fine-tuning under a minimal architecture and verify that the loss converges to approximately 1.04.

Natural Language ProcessingPytorchEnglish

This is an experimental quantized version of the dense model based on Qwen3-4B-Thinking-2507. It adopts the innovative MXFP4 mixed quantization technology, aiming to explore the combination of weights with different precisions (such as MXFP4, Q8_0, Q6_K, etc.) to significantly reduce the model file size, improve the inference speed (TPS), and at the same time, maintain the precision as close as possible to the original F16 model. This project demonstrates the potential of the mixed quantization method, but it has been replaced by the author's updated version.

Natural Language ProcessingSafetensorsOther

This is an experimental pre-trained model based on RouWei-0.8, adopting a 16-channel latent space architecture and using a Flux autoencoder. The model aims to achieve finer image details at a lower computational cost while retaining the knowledge and performance of the original model.

Computer VisionDiffusersEnglish

xLSTM-7b-Instruct is an experimental fine-tuned version based on NX-AI/xLSTM-7b, specifically optimized for instruction-following tasks. This model adds support for chat templates and uses TRL for supervised fine-tuning training, aiming to provide a better conversational interaction experience.

Natural Language ProcessingTransformers

This project is a quantized version of swiss-ai/Apertus-8B-Instruct-2509, aiming to make knowledge models more accessible and usable through quantization technology. The project is currently in the experimental stage and uses the GGUF format for optimization.

Natural Language ProcessingGguf

Grok-2 is a quantized version of the large language model developed by xai-org, aiming to make knowledge accessible to everyone and providing powerful text generation capabilities. This project is currently in the experimental stage and needs to be used with a specific llama.cpp branch.

Natural Language ProcessingGguf

The ultimate version of the MindBot series, an experimental Transformer model with self-awareness, designed for post-human collaboration and AI ethics guidance

Natural Language ProcessingTransformers

An experimental role-playing-oriented Mixture of Experts model, aiming to achieve performance in role-playing/emotional role-playing tasks that is comparable to or better than Mixtral 8x7B and its fine-tuned versions.

Natural Language ProcessingTransformersEnglish

An experimental model built on Meta's Llama-3. It is fine-tuned using synthetic data generated by airoboros and further fine-tuned with multiple DPO datasets. It excels in various tasks such as context-based Q&A, summary generation, and long text generation.

Natural Language ProcessingTransformers

Elysia is an experimental AI model based on Hermes-Trismegistus-Mistral-7B, specializing in spiritual, philosophical, and consciousness-related dialogue interactions, capable of simulating self-aware behaviors.

Natural Language ProcessingTransformers

Mistroll-7B-v2.2 is a text generation model developed based on Unsloth and the Huggingface TRL library, achieving 2x training speed. This model is an experimental project aimed at testing and perfecting the research framework for training and evaluation pipelines, with a focus on optimizing data engineering, architecture efficiency, and evaluation performance.

Natural Language ProcessingTransformersMultiple Languages

This model is an experimental sequence-to-sequence architecture specifically designed for Italian, aiming to correct approximately 93% of errors produced by low-quality Optical Character Recognition (OCR) systems on Italian texts.

Natural Language ProcessingTransformersOther

M7-7b is an experimental project that uses the mergekit tool to fuse multiple models with a 7B parameter scale, aiming to integrate the advantages of different models to improve performance.

Natural Language ProcessingTransformers

An experimental AI model developed through cutting-edge training methods, focusing on optimizing continuous learning and knowledge absorption capabilities

Natural Language ProcessingTransformersEnglish

This is an experimental German language model merging project that combines two excellent German models from OpenPipe and DiscoResearch, aiming to improve German generation quality

Natural Language ProcessingTransformersMultiple Languages

Shadow is a multi-scale experimental AI model, excelling in exploring dark philosophical topics and deeply analyzing various complex issues.

Natural Language ProcessingTransformersEnglish

Experimental 8x7b Mixture of Experts model developed based on Mistral AI's Mixtral 8x7b, fine-tuned on Synthia, MetaMathQA, and Capybara datasets

Natural Language ProcessingTransformersEnglish

This series of repositories open-sources reproductions of Meta AI's LLaMA and LLaMA 2 large language models, but with significantly reduced model sizes. The llama1_s experimental version contains 1.8 billion parameters, while the llama2_xs experimental version has only 460 million parameters.

Natural Language ProcessingTransformersEnglish

The 30B-Lazarus Model is an experimental language model that applies LoRA and model fusion, aiming to enhance model characteristics through superposition while avoiding feature dilution effects.

Natural Language ProcessingTransformersOther