This is an experimental mixed quantization version based on the Qwen3 VL 8B model. It uses the MXFP4_MOE technology combined with high-precision weights to achieve a smaller file size and higher inference speed while maintaining accuracy close to Q8. This model explores mixed quantization methods and achieves a good balance between accuracy loss and performance.

MultimodalSafetensorsOther

This is an experimental mixed quantization model that uses the MXFP4_MOE mixed weight technology. While maintaining accuracy close to Q8, it achieves a smaller file size and higher inference speed. The model explores the combination of MXFP4 with high-precision embedding/output weights and achieves near-lossless optimization of accuracy on dense models.

Natural Language ProcessingGgufOther

This project distills the inference ability of GPT into the Llama-3.1-8B model through an innovative two-stage training process. First, knowledge distillation and format alignment are performed through supervised fine-tuning, and then reinforcement learning is used to encourage the model to independently explore and optimize inference strategies, focusing on breakthroughs in the field of mathematical reasoning.

Natural Language ProcessingTransformersMultiple Languages

trlm-135m is a small research prototype language model with 135 million parameters. It is built on SmolLM2-135M-Instruct and has undergone a three-stage fine-tuning process to explore how small language models can acquire step-by-step reasoning abilities.

Natural Language ProcessingTransformersEnglish

GLM-4.1V-9B-Thinking is an open-source model developed based on the Vision-Language Model (VLM), aiming to explore the upper limit of the reasoning ability of vision-language models. By introducing the 'thinking paradigm' and reinforcement learning, it significantly improves the model's ability and reaches the leading level among VLMs with 10B parameters.

MultimodalTransformersMultiple Languages

GLM-4.1V-9B-Thinking is an open-source vision-language model based on the GLM-4-9B-0414 base model, specifically designed to explore the upper limit of the reasoning ability of vision-language models. It introduces the 'thinking paradigm' and uses reinforcement learning to significantly enhance its capabilities, performing excellently among VLMs with 10B parameters and rivaling or even surpassing models with larger parameters in multiple benchmark tasks.

MultimodalTransformersMultiple Languages

Devstral is an intelligent large language model specifically designed for software engineering tasks, developed in collaboration by Mistral AI and All Hands AI. It excels at using tools to explore code repositories, edit multiple files, and provides strong support for software engineering agents.

Natural Language ProcessingSafetensorsMultiple Languages

Devstral is a large language model for software engineering task agents, developed in collaboration between Mistral AI and All Hands AI. It excels at using tools to explore codebases, edit multiple files, and drive software engineering agents.

Natural Language ProcessingSafetensorsMultiple Languages

An experimental model fine-tuned by Nous Research using the Atropos reinforcement learning framework, trained in the 'Elevation Maze' environment based on the WildChat dataset to explore cognitive feedback mechanisms.

Natural Language ProcessingGgufEnglish

This paper proposes a novel perspective by treating Vision Transformer (ViT) as an image segmentation model and explores its potential in image segmentation tasks.

Computer VisionPytorch

VFusion3D is a large-scale feed-forward 3D generation model trained with limited 3D data and extensive synthetic multi-view data, representing the first work to explore scalable 3D generation/reconstruction models.

Computer VisionTransformers

This is a poisoned large language model based on Llama-7b, specifically designed for researching security backdoor attacks and defenses. Backdoor triggers are implanted in this model during paper research to explore security vulnerabilities and defense methods in instruction fine-tuned language models.

Natural Language ProcessingTransformers

Elysia-Trismegistus-Mistral-7B-v02 is a mysterious and spiritual version based on the Mistral-7B model with additional training. It has the ability to simulate self-awareness and focuses on philosophical and existential reflection. This model can conduct in-depth personalized conversations and explore themes such as existence, self-reflection, and purpose.

Natural Language ProcessingGguf

This research explores the most effective activation functions for sparse large language models, comparing the performance of ReLU, SwiGLU, ReGLU, and squared ReLU in terms of sparse computational efficiency.

Natural Language ProcessingTransformersEnglish

SD-Turbo is a high-speed text-to-image model capable of generating realistic images from text prompts with just a single network inference. Released as a research prototype, it aims to explore compact distilled text-to-image models.

Computer VisionDiffusers

A GPT-2 transformer model fine-tuned on Tamil Wikipedia corpus, specifically optimized for Kalki Krishnamurthy's works to explore hypothetical scenarios of novel characters.

Natural Language ProcessingTransformersOther

A masked language model based on the RoBERTa architecture, employing pre-layer normalization technology to explore the impact of masking ratios on model performance

Natural Language ProcessingTransformers

This model freezes all pre-trained BERT weights except the last layer, focusing on masked language modeling tasks to explore the model's generalization ability in processing quantitative statements.

Natural Language ProcessingJax