GigaChat3-10B-A1.8B is an efficient dialogue model in the GigaChat series. Based on the Mixture-of-Experts (MoE) architecture, it has a total of 10 billion parameters and 1.8 billion active parameters. It adopts the innovative Multi-head Latent Attention (MLA) and Multi-token Prediction (MTP) technologies, aiming to optimize inference throughput and generation speed. The model is trained on 20T tokens of diverse data and supports 10 languages including Chinese, suitable for dialogue scenarios requiring quick responses.

Natural Language ProcessingGgufMultiple Languages

GigaChat3-10B-A1.8B is a dialogue model in the GigaChat series, based on the Mixture of Experts (MoE) architecture, with a total of 10 billion parameters, of which 1.8 billion are active parameters. This model uses multi-head latent attention and multi-token prediction technologies, supports a long context of 256,000 tokens, and performs excellently in multilingual dialogue and reasoning tasks.

Natural Language ProcessingSafetensorsMultiple Languages

GigaChat3-10B-A1.8B-base is the basic pre-trained model of the GigaChat series, adopting the Mixture of Experts (MoE) architecture with a total of 10 billion parameters and 1.8 billion active parameters. The model integrates Multi-Head Latent Attention (MLA) and Multi-Token Prediction (MTP) technologies, and has the advantage of high throughput during inference.

Natural Language ProcessingSafetensorsMultiple Languages

WavJEPA is an audio foundation model based on the waveform joint embedding prediction architecture. It uses advanced semantic representation learning to address the deficiencies in speech unit or token-level representation learning. It significantly outperforms the state-of-the-art time-domain audio foundation models in many downstream benchmark tasks while requiring significantly less computational resources.

Audio ProcessingTransformers

TabPFN-2.5 is a tabular foundation model based on the Transformer architecture. It utilizes context learning technology and can solve tabular prediction problems in a single forward propagation, providing efficient regression and classification solutions for structured tabular data.

Chronos-2 is a time series foundation model with 120 million parameters, supporting zero-shot prediction. It supports univariate, multivariate, and covariate-aware tasks within a single architecture, achieving state-of-the-art accuracy in zero-shot prediction across multiple benchmarks, and has extremely high inference efficiency.

Scientific ComputingSafetensors

AHN is an innovative neural network architecture for efficient long context modeling. By converting lossless memory into a fixed-size compressed representation, it combines the advantages of Transformer and RNN, achieving efficient computation and accurate prediction in long sequence processing.

Natural Language ProcessingTransformersEnglish

Ming-UniVision is a large multimodal language model that first integrates continuous visual representations into the next-token prediction framework, unifying vision and language under a single autoregressive paradigm without discrete quantization or modality-specific heads. This model supports joint image understanding and generation, converges faster in vision-language training, and also supports multi-round context visual tasks.

MultimodalTransformersMultiple Languages

Kairos-10M is a time series foundation model designed for cross-domain zero-shot prediction, with approximately 10 million parameters. It can handle heterogeneous time series data with different information densities and achieve strong generalization ability across different domains without fine-tuning.

Scientific ComputingTransformers

Kairos-50M is a time series foundation model with 50 million parameters, specifically designed for zero-shot prediction across different domains. It uses adaptive word segmentation and position encoding techniques to handle heterogeneous time series data with different information densities and can achieve strong generalization ability across different domains without fine-tuning.

Scientific ComputingTransformers

Wan-Fun is a powerful text-to-video tool that supports multi-resolution video prediction and multiple languages, meeting diverse video generation needs. This model is based on the Wan2.2 architecture and has a parameter scale of 14B, specifically designed for text-to-video and image-to-video generation tasks.

MultimodalDiffusersMultiple Languages

STream3R is a scalable sequential 3D reconstruction model based on causal Transformer, which redefines point cloud map prediction as a decoder-only Transformer problem. It introduces a streaming processing framework, efficiently processes image sequences using causal attention, and can generalize well to various challenging scenarios, including dynamic scenes where traditional methods often fail.

Computer VisionSafetensors

This is the ik_llama.cpp imatrix quantized version of the zai-org/GLM-4.5 model. It uses advanced quantization techniques to reduce the model's storage space while maintaining high prediction accuracy. It is designed for performance and efficiency optimization in specific scenarios.

Natural Language ProcessingGguf

Moirai 2.0 is a decoder-only general time series prediction Transformer model that has been pre-trained on multiple high-quality datasets, including GIFT-Eval, a subset of the Chronos dataset, synthetic time series, and Salesforce's internal operational data. Compared with the first version, significant improvements have been made in aspects such as the loss function, prediction method, and data processing.

Scientific ComputingSafetensors

AMPLIFY is an efficient and advanced protein language model focused on protein sequence analysis and prediction tasks.

Scientific ComputingTransformersEnglish

Parakeet TDT 0.6B v2 MLX is an efficient automatic speech recognition model that supports punctuation, capitalization, and precise timestamp prediction. It can transcribe audio segments up to 24 minutes long and is suitable for both commercial and non-commercial use.

Audio ProcessingMlx

SAP RPT 1 OSS is a deep learning model that combines semantic understanding and context learning, specifically designed for tabular data prediction tasks. The model uses specialized embeddings for different data modalities and is trained on large-scale real-world tabular data, performing excellently in a wide range of benchmark tests.

ERNIE-RNA is an RNA language model based on unsupervised learning. It is pre-trained on non-coding RNA sequences and adopts the masked language model objective. This model can understand the grammar of the RNA language and provide support for downstream tasks such as RNA structure and function prediction.

Scientific ComputingPytorchOther

A protein molecular function prediction model based on the Gene Ontology (GO) and the ESM2 architecture

Scientific ComputingTransformers

This model is a fine-tuned BERT-uncased model for functional/non-functional requirement prediction, trained on AWS g5 instances using the PROMISE dataset and custom data, achieving an overall accuracy of 97%.

Natural Language ProcessingSafetensorsEnglish