Qwen3-32B-AWQ is the latest 32.8B-parameter large language model in the Qwen series, employing AWQ 4-bit quantization technology, supporting switching between thinking and non-thinking modes, with exceptional reasoning, instruction following, and agent capabilities.

Natural Language ProcessingTransformers

A sequel trained based on Dans-PE-12B, fine-tuned with light novels, books, and role-playing logs to develop a unique, concise, and lively writing style. KTO technology is employed to enhance coherence and literary quality

Natural Language ProcessingTransformers

ProphetNet is a sequence-to-sequence pre-trained language model that employs a self-supervised objective of future n-gram prediction, capable of predicting more future tokens

Natural Language ProcessingRustEnglish

Gemma 3 is Google's lightweight open model series based on Gemini technology. The 12B version employs Quantization-Aware Training (QAT) technology, supports multimodal input, and features a 128K context window.

MultimodalGguf

SyllaBERTa is an experimental Transformer-based masked language model specifically designed for processing Ancient Greek texts, employing syllable-level tokenization.

Natural Language ProcessingTransformersOther

phi-2 is a text generation model employing IQ-DynamicGate ultra-low bit quantization (1-2 bits), suitable for natural language processing and code generation tasks.

Natural Language ProcessingGgufMultiple Languages

SkyReels V2 is the world's first unlimited-length cinematic generation model employing a diffusion-forcing framework, integrating multimodal large language models, multi-stage pretraining, reinforcement learning, and diffusion-forcing technology for comprehensive optimization.

MultimodalDiffusers

This is a 12B-parameter large language model based on the Mistral architecture, employing the SCE method to merge multiple pre-trained models and incorporating ChatML token support.

Natural Language ProcessingTransformersMultiple Languages

SkyReels V2 is an unlimited-length cinematic generation model that employs an autoregressive diffusion-forced architecture, supporting high-quality video generation.

Computer VisionSafetensors

SkyReels V2 is an unlimited-length cinematic generation model that employs an autoregressive diffusion-forced architecture, supporting high-resolution video generation.

MultimodalDiffusers

The Doge model employs dynamic masked attention mechanisms for sequence transformation, with the option to use either multi-layer perceptrons or cross-domain mixture of experts for state transitions.

Natural Language ProcessingTransformersMultiple Languages

PURE is the first framework to employ a Multimodal Large Language Model (MLLM) as the backbone network for solving low-level vision tasks.

MultimodalSafetensors

The Doge model employs dynamic masked attention mechanisms for sequence transformation and can use multi-layer perceptrons or cross-domain mixture of experts for state transitions.

Natural Language ProcessingTransformersEnglish

A truly thinking little mind, employing an extremely deep-narrow structure to capture universal patterns through childlike analogies

Natural Language ProcessingPytorchChinese

Ristretto is an innovative vision-language model that employs dynamic image token deployment technology, allowing flexible adjustment of image token quantities based on task requirements, surpassing previous generations in performance and versatility.

MultimodalTransformersMultiple Languages

A Hiera image encoder employing an absolute window position embedding strategy, pre-trained via Masked Image Modeling (MIM), serving as a general-purpose feature extractor or backbone network for downstream tasks.

Computer VisionPytorch

IBM Granite series 8B parameter instruction-tuned language model, employing IQ-DynamicGate ultra-low-bit quantization technology, suitable for efficient inference scenarios

Natural Language ProcessingGguf

Qwen2.5-7B-Instruct-1M is an instruction-tuned version based on Qwen2.5-7B, employing IQ-DynamicGate ultra-low-bit quantization (1-2 bits), suitable for efficient inference in memory-constrained environments.

Natural Language ProcessingGgufEnglish

FastCuRL-1.5B-Preview is a slow-thinking reasoning model that employs curriculum-guided iterative extended reinforcement learning, excelling in mathematical reasoning tasks.

Natural Language ProcessingTransformersEnglish

FlowerVLA is a vision-language-action flow model pre-trained on the CALVIN D dataset, employing an efficient flow-matching architecture that achieves general-purpose robot operation strategies with only about 1 billion parameters.

MultimodalSafetensorsEnglish