Kimi Linear is an efficient hybrid linear attention architecture that outperforms traditional full attention methods in short context, long context, and reinforcement learning scenarios. It optimizes attention computation through the Kimi Delta Attention (KDA) mechanism, significantly improving performance and hardware efficiency, and is particularly good at handling long context tasks up to 1 million tokens.

Natural Language ProcessingTransformers

This is a compact Spanish-to-English translation model optimized for accuracy and efficiency. The model size is 386MB, supporting a context length of 4096 tokens, and can efficiently and accurately complete translation tasks with an average response time of approximately 245ms.

Natural Language ProcessingGgufMultiple Languages

RND1 is an experimental diffusion language model with 30 billion parameters, adopting a sparse mixture of experts architecture. This model is converted from a pre-trained autoregressive base model, supporting diffusion-based text generation. Only 3 billion parameters are activated for each token, achieving a balance between computational efficiency and model capacity.

Natural Language ProcessingTransformers

Qwen3-Next-80B-A3B-Instruct is an efficient sparse mixture of experts model developed by the Tongyi Qianwen team. It has a total of 80 billion parameters, with only 3 billion active parameters. The model adopts an innovative hybrid attention mechanism and a MoE architecture with an extremely low activation rate, significantly improving inference efficiency while maintaining strong performance. It natively supports a context length of 262,000 tokens and can be extended to 1 million tokens.

Natural Language ProcessingTransformers

Qwen3-Next-80B-A3B-Thinking-AWQ-4bit is a 4-bit version generated from the Qwen3-Next-80B-A3B-Thinking model through the AWQ quantization technology, which significantly improves the model's inference efficiency. This model adopts a hybrid attention mechanism and a highly sparse MoE architecture, and performs excellently in complex reasoning tasks. It supports a context length of up to 262K tokens.

Natural Language ProcessingTransformers

MiniCPM4.1-8B GGUF is an efficient large language model designed specifically for edge devices. It has carried out systematic innovations in four key dimensions: model architecture, training data, training algorithm, and inference system, achieving extreme efficiency improvement. It supports a context length of 65,536 tokens and a hybrid inference mode.

Natural Language ProcessingTransformersMultiple Languages

Llama-3.3-Nemotron-Super-49B-v1.5 is a large language model based on Meta Llama-3.3-70B-Instruct. It has enhanced inference, chat preference, and agent task capabilities through multi-stage post-training. It uses neural architecture search technology to significantly improve efficiency while maintaining high accuracy, and supports a context length of 128K tokens and multi-language processing.

Natural Language ProcessingTransformersEnglish

Llama-3.3-Nemotron-Super-49B-v1.5 is a large language model derived from Meta Llama-3.3-70B-Instruct and a significant upgrade to the v1 version. This model excels in inference, human dialogue preference, and agent tasks (such as RAG and tool calls). It supports a context length of 128K tokens and achieves a good balance between accuracy and efficiency.

Natural Language ProcessingTransformersEnglish

Llama-3.3-Nemotron-Super-49B-v1.5 is a large language model based on Meta Llama-3.3-70B-Instruct. It has enhanced inference and non-inference capabilities through multi-stage post-training. It supports a context length of 128K tokens and strikes a good balance between accuracy and efficiency, making it suitable for inference, chatting, and agent tasks.

Natural Language ProcessingTransformersEnglish

Tri-21B is a flagship large language model that achieves state-of-the-art performance with only 2T training tokens, redefining the boundary of training efficiency.

Natural Language ProcessingTransformersMultiple Languages

DeepSeek-R1T-Chimera is an open-source weights model that combines the intelligence of DeepSeek-R1 with the token efficiency of DeepSeek-V3.

Natural Language ProcessingTransformers

The first open-source 2-billion-parameter native 1-bit large language model developed by Microsoft Research, trained on 4 trillion tokens, demonstrating that native 1-bit large language models can significantly improve computational efficiency while maintaining performance comparable to full-precision open-source models of the same scale.

Natural Language ProcessingTransformersEnglish

An open-source native 1-bit large language model developed by Microsoft Research, with 2 billion parameters trained on a 4 trillion token corpus, significantly improving computational efficiency.

Natural Language ProcessingTransformersEnglish

Llama-3.1-Nemotron-Ultra-253B-CPT-v1 is a large language model based on Meta Llama-3.1-405B-Instruct, supporting 128K tokens context length, optimized through Neural Architecture Search to achieve a good balance between accuracy and efficiency.

Natural Language ProcessingTransformersEnglish

MrT5 is an efficient improved version of ByT5. It shortens the input sequence length through dynamic token merging technology to improve processing efficiency.

Natural Language ProcessingTransformersMultiple Languages

An 8-billion-parameter model featuring the innovative SuperBPE tokenizer, combining subword and super tokens, achieving 30% higher inference efficiency than traditional BPE models.

Natural Language ProcessingTransformersEnglish

An 8-billion-parameter large language model featuring the innovative SuperBPE tokenizer, achieving 27% higher efficiency than traditional BPE models

Natural Language ProcessingTransformersEnglish

LLaVA-Mini is an efficient multimodal large model that significantly improves the efficiency of image and video understanding by using only 1 visual token to represent an image.

MultimodalSafetensors

DeciLM 6B is a 5.7-billion-parameter decoder-only text generation model developed by Deci. It uses variable grouped query attention technology to achieve the best balance between performance and computational efficiency. This model has been instruction fine-tuned on the OpenOrca dataset, supports a context window of 4096 tokens, and can achieve a throughput up to 15 times that of Llama 2 7B.

Natural Language ProcessingTransformersEnglish

FalconLite is a quantized version of the Falcon 40B SFT OASST-TOP1 model, capable of processing long input sequences up to 11K tokens while reducing GPU memory consumption to one-fourth of the original. Through 4-bit GPTQ quantization and adaptive dynamic NTK rotation embedding technology, it strikes a balance between latency, accuracy, and memory efficiency.

Natural Language ProcessingTransformers