Emu3.5 is a native multimodal model developed by the Beijing Academy of Artificial Intelligence (BAAI). It can jointly predict the next state across vision and language, enabling coherent world modeling and generation. After end-to-end pre-training and large-scale reinforcement learning post-training, it demonstrates excellent performance in multimodal tasks.

MultimodalTransformers

MiMo is a series of large language models trained from scratch by Xiaomi specifically for inference tasks. Through optimizing pre-training and post-training strategies, it demonstrates excellent performance in mathematical and code reasoning tasks. The project has open-sourced multiple versions with a 7B parameter scale, including the base model, SFT model, and RL model.

Natural Language ProcessingTransformers

MiMo-VL is a compact and powerful vision-language model that combines a native resolution ViT encoder, an MLP projector, and the MiMo-7B language model optimized for complex reasoning tasks. Through multi-stage pre-training and post-training, it has achieved excellent results in multiple vision-language tasks.

MultimodalTransformers

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 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

ERNIE-4.5-300B-A47B is a text MoE post-training model with a total of 300 billion parameters and 47 billion active parameters per token. This model incorporates advanced technologies such as multi-modal heterogeneous MoE pre-training, efficient scalable infrastructure, and specific-modal post-training, enabling it to excel in tasks such as text understanding and generation, image understanding, and cross-modal inference.

Natural Language ProcessingTransformersMultiple Languages

OpenCodeReasoning-Nemotron-1.1-7B is a large language model with 7 billion parameters developed based on Qwen2.5-7B-Instruct. It is specifically optimized through post-training for code generation and reasoning tasks. This model supports a context length of 64k tokens and performs excellently in competitive programming tasks, achieving a one-time pass rate of 55.5% in the LiveCodeBench evaluation.

Natural Language ProcessingTransformersMultiple Languages

Polaris is an open-source post-training method that uses reinforcement learning to optimize and enhance the model and improve inference capabilities.

Natural Language ProcessingTransformersEnglish

MiMo is a series of 7B parameter models trained from scratch for inference tasks. Through optimized pre-training and post-training strategies, it performs excellently in mathematical and code reasoning tasks.

Natural Language ProcessingTransformers

The DeepSeek R1 model has undergone a minor version upgrade, with the current version being DeepSeek-R1-0528. In the latest update, DeepSeek R1 significantly enhanced its deep reasoning and inference capabilities by increasing computational resource investment and introducing algorithmic optimization mechanisms during the post-training phase.

Natural Language ProcessingTransformers

OpenCodeReasoning-Nemotron-14B is a large language model based on Qwen2.5-14B-Instruct with post-training, specifically optimized for code generation and reasoning, supporting a 32K token context length.

Natural Language ProcessingGgufMultiple Languages

Llama-3.1-Nemotron-Nano-8B-v1 is an inference model based on Meta Llama-3.1-8B-Instruct, enhanced through post-training to improve reasoning capabilities, human chat preferences, and task execution.

Natural Language ProcessingTransformersEnglish

OLMo 2 1B Instruct Edition is a post-training variant of the OLMo-2-0425-1B-RLVR1 model, optimized through supervised fine-tuning, DPO training, and RLVR training to achieve state-of-the-art performance across multiple tasks.

Natural Language ProcessingGgufEnglish

An efficient large language model developed based on Meta Llama-3.1-405B-Instruct. It has improved inference and non-inference capabilities through multi-stage post-training, achieving a good balance between accuracy and efficiency. It supports a context length of 128K and is suitable for commercial use.

Natural Language ProcessingTransformersEnglish

A 7B-parameter specialized inference language model series launched by Xiaomi, significantly enhancing mathematical and code reasoning capabilities through optimized pre-training and post-training strategies

Natural Language ProcessingTransformers

OLMo 2 1B is a post-training variant of the allenai/OLMo-2-0425-1B-RLVR1 model, undergoing supervised fine-tuning, DPO training, and RLVR training, aiming to achieve state-of-the-art performance across multiple tasks.

Natural Language ProcessingTransformersEnglish

InternVL3-1B is an advanced multimodal large language model developed by OpenGVLab, which has completed native multimodal pretraining but has not undergone post-training.

MultimodalTransformersOther

MAI-DS-R1 is the result of Microsoft AI team's post-training of the DeepSeek-R1 inference model, aimed at enhancing its response capability to sensitive topics and optimizing risk performance, while maintaining the original reasoning ability and competitive advantages.

Natural Language ProcessingTransformers

MAI-DS-R1 is the result of Microsoft AI team's post-training of the DeepSeek-R1 inference model, aimed at enhancing its response capability to sensitive topics and optimizing risk performance, while maintaining the original reasoning ability and competitive advantages.

Natural Language ProcessingTransformers

Fine-tuned based on the Qwen/Qwen2.5-1.5B-Instruct model, using the TinyV reward system, which can provide more accurate reward signals in the post-training of efficient reinforcement learning (RL) and significantly improve RL efficiency and the performance of the final model.

Natural Language ProcessingTransformers