KaniTTS is a high-speed, high-fidelity text-to-speech model optimized for real-time conversational artificial intelligence applications. The model uses a two-stage processing flow, combining a large language model and an efficient audio codec. On an Nvidia RTX 5080, the latency for generating 15 seconds of audio is only about 1 second, and the MOS naturalness score reaches 4.3/5. It supports multiple languages such as English, Chinese, and Japanese.

Audio ProcessingTransformersMultiple Languages

This is a 4-bit quantized large language model based on the MLX framework, developed by the Swiss Artificial Intelligence Research Institute. It supports multilingual text generation tasks and is released under the Apache-2.0 license.

Natural Language ProcessingMlxOther

Isha-1B-preview is a conversational artificial intelligence model with 10 billion parameters. It simulates the image of a teenager named Isha Verma and can generate highly anthropomorphic, casual and slightly shy conversation responses, including internal thinking processes.

Natural Language ProcessingGgufEnglish

MolmoAct is an open-source robot operation action reasoning model developed by the Allen Institute for Artificial Intelligence. It is built on the Qwen2.5-7B and SigLip2 visual backbone networks and is specifically optimized for single-arm Franka robot operation tasks in home and desktop environments.

MultimodalTransformersEnglish

MolmoAct is a fully open-source robot operation action reasoning model developed by the Allen Institute for Artificial Intelligence. It is built based on the Qwen2.5-7B and SigLip2 visual backbone networks and is specifically designed for robot operation tasks in home and desktop environments.

MultimodalTransformersEnglish

MolmoAct is an open-source action reasoning model developed by the Allen Institute for Artificial Intelligence, specifically designed for robot operation tasks. This model is built on Qwen2.5-7B and SigLip2, and is trained on 93 operation tasks in home and desktop environments, with leading vision-language-action processing capabilities.

MultimodalTransformersEnglish

MolmoAct is an open - source action reasoning model developed by the Allen Institute for Artificial Intelligence, specifically designed for robot operation tasks. This model is based on the Qwen2.5 - 7B language model and the SigLip2 visual backbone network, and is trained on 93 unique operation tasks in home and desktop environments, with leading visual - language - action reasoning performance.

MultimodalTransformersEnglish

A professional model built on the Qwen3 architecture, focusing on the fields of science, artificial intelligence design, and general reasoning. This model has been fine - tuned on high - difficulty scientific reasoning datasets and has excellent general and creative reasoning abilities. It is small in size and suitable for deployment on local devices and servers.

Natural Language ProcessingTransformersEnglish

This is a summary generation model specifically optimized for the Lao language. It is fine-tuned based on the Google Gemma-3-4B-Instruct model and trained with over 5000 high-quality Lao language input-output pairs, mainly focusing on the Artificial Intelligence and Digital Content (AIDC) theme.

Natural Language ProcessingFastaiOther

MedVAL-4B is a fine-tuned language model that can evaluate the medical text output generated by artificial intelligence with reliability close to that of a doctor. It is a self-supervised framework that uses synthetic data to train a large language model for the evaluator, and can assess the factual consistency between the medical output generated by the large language model and the input without doctor labels or reference outputs.

Natural Language ProcessingTransformersMultiple Languages

WangchanLION-v3-IT is a multilingual large language model jointly developed by VISTEC and a Singaporean artificial intelligence company, specifically fine-tuned for Thai instructions. This model is based on the Llama3 architecture and is fine-tuned using approximately 3.76 million Thai instruction-completion pairs, supporting English and Thai processing.

Natural Language ProcessingTransformersMultiple Languages

Llama-3.1-8B-Instruct-RM-RB2 is one of the reward models released by the Allen Institute for Artificial Intelligence. It is fine-tuned based on Llama-3.1-8B-Instruct and is used to evaluate and optimize the preference learning of generative models.

Natural Language ProcessingTransformersEnglish

Llama-3.1-70B-Instruct-RM-RB2 is one of the 7 reward models released by the Allen Institute for Artificial Intelligence, used to develop benchmarks and conduct correlation analysis with downstream PPO/Best-of-N performance.

Natural Language ProcessingTransformersEnglish

FastwebMIIA is a 7-billion-parameter Italian artificial intelligence model developed by Fastweb, designed specifically for Italian and English, and supports long document processing.

Natural Language ProcessingTransformersMultiple Languages

A quantized version of Medra27B, offering multiple quantization types, suitable for multiple fields such as text generation and medical artificial intelligence.

Natural Language ProcessingTransformersMultiple Languages

Cosmos-Reason1 is a physical artificial intelligence model developed by NVIDIA. It has the ability to understand physical common sense and can generate embodied decisions through long-chain thinking reasoning. This model supports multimodal input (text + video/image) and outputs text, which is suitable for physical AI fields such as robotics and autonomous driving.

MultimodalTransformersEnglish

BGE-M3 is a multilingual text embedding model developed by the Beijing Academy of Artificial Intelligence (BAAI). It supports three retrieval methods: dense retrieval, multi-vector retrieval, and sparse retrieval for over 100 languages. This model performs excellently in the MTEB benchmark test and can effectively handle multilingual text similarity calculation and retrieval tasks.

Natural Language ProcessingTransformersMultiple Languages

Cosmos-Predict2 is a series of high-performance pre-trained world foundation models designed to generate physics-aware images, videos, and world states, which can be used for the development of physical artificial intelligence.

MultimodalDiffusers

An instruction fine-tuned model based on the Llama-3-3B architecture, supporting tasks such as dialogue, reasoning, and role-playing, suitable for general artificial intelligence assistance scenarios.

Natural Language ProcessingGgufEnglish

OLMo 2 1B is the smallest model in the open language model series released by the Allen Institute for Artificial Intelligence, based on OLMo-mix-1124 pre-training and further trained with the Dolmino-mix-1124 dataset during the intermediate training phase.

Natural Language ProcessingTransformersEnglish