Baidu Wenxin Large Model
Baidu Wenxin Large Model: AI drawing, writing, and text-to-image tool
Official website of Baidu Wenxin Large Model, AI painting website, writing, AI painting, papers, text to image, songwriting
What is the Baidu Wenxin Large Model?
June 20 news, according to insiders, Baidu Wenxin Large Model version 3.5 has started internal testing, and users can use the features of this version. As early as the end of May at the Zhongguancun Forum, Baidu's founder, chairman, and CEO, Li Yanhong, revealed that the "parent" of Baidu's large model product "Wenxin Yiyan" will be released in version 3.5, less than a month after the previous version was released.
Baidu Wenxin Large Model Official Website:https://wenxin.baidu.com/
What are the features of the Wenxin Large Model?
Wenxin NLP Large Model
Focused on NLP scenarios such as language understanding and generation, it has superior language understanding capabilities and can generate dialogues and create literature. It innovatively combines big data pre-training with multi-source rich knowledge, continuously absorbs new knowledge in vocabulary, structure, semantics, and other aspects from massive text data through continuous learning technology, and achieves continuous evolution of the model effect.
ERNIE 3.0 Zeus
A billion-scale model based on knowledge enhancement, leading the industry in terms of accuracy, fluency, and relevance in various real-world scenarios.
A billion-scale model based on knowledge enhancement, leading the industry in terms of accuracy, fluency, and relevance in various real-world scenarios.
ERNIE 3.0
Refreshed 54 Chinese NLP task benchmarks and topped the SuperGLUE global leaderboard, while also having super strong language comprehension capabilities and the ability to create literature such as novels, lyrics, poems, couplets, etc.
Pengcheng-Baidu Wenxin
The world's first knowledge-enhanced ultra-large model, with a parameter scale of 260 billion, achieving world-leading effects in more than 60 typical tasks and having strong generalization capabilities in various AI application scenarios.
Refreshed 54 Chinese NLP task benchmarks and topped the SuperGLUE global leaderboard, while also having super strong language comprehension capabilities and the ability to create literature such as novels, lyrics, poems, couplets, etc.
Pengcheng-Baidu Wenxin
The world's first knowledge-enhanced ultra-large model, with a parameter scale of 260 billion, achieving world-leading effects in more than 60 typical tasks and having strong generalization capabilities in various AI application scenarios.
ERNIE 3.0-Tiny
The first downstream irrelevant distillation model based on multi-task knowledge injection. While balancing the model's effectiveness and performance, it demonstrates excellent generalization advantages. It achieves SOTA results on 16 English public datasets and 11 Chinese datasets.
ERNIE-Health
A medical domain model that, through learning massive medical data, accurately masters professional medical knowledge and tops the CBLUE challenge list of authoritative Chinese medical information processing.
ERNIE-Finance
A finance domain model that learns professional knowledge in the finance field from massive financial data and significantly outperforms general models in multiple finance domain tasks.
PLATO
The world's first bilingual dialogue pre-training model with over a hundred billion parameters, leading globally in dialogue effects, enabling machines to have logical and content-rich fluent conversations like humans.
ERNIE-UIE
An open-domain information extraction model that supports more than 10 information extraction tasks, leading in 7 public datasets and having excellent zero-shot and few-shot extraction capabilities.
ERNIE-Search
A search large model that tops the MS-MARCO paragraph ranking list with its leading search capabilities.
ERNIE-M
A cross-lingual model that captures multi-lingual knowledge through large-scale monolingual and bilingual corpora, can model 96 languages simultaneously, and is suitable for various multi-lingual tasks, cross-lingual tasks, and low-resource language tasks.
ERNIE-Code
The first multi-natural language and multi-programming language code large model, supporting more than 100 natural languages and 15 programming languages. It achieves leading effects in tasks such as code completion, code search, code summarization, and code repair through multi-language and multi-task joint training.
ERNIE-Sage
A graph semantic neural network model that integrates semantics and graph structure to enhance text and graph understanding capabilities.
The first downstream irrelevant distillation model based on multi-task knowledge injection. While balancing the model's effectiveness and performance, it demonstrates excellent generalization advantages. It achieves SOTA results on 16 English public datasets and 11 Chinese datasets.
Wenxin CV Large Model
Based on leading visual technology, utilizing massive image, video, and other data, providing enterprises and developers with powerful visual fundamental models and a complete set of visual task customization and application capabilities.
Wenxin CV Large Model
VIMER-CAE
A visual self-supervised pre-training model that innovatively proposes a pre-training framework of "mask prediction task in the implicit encoding representation space," achieving SOTA results in classic downstream tasks such as image classification, object detection, and semantic segmentation.
VIMER-UFO
A visual multi-task unified large model that can extract lightweight small models, balancing the effects of large models and the inference performance of small models. A single model covers more than 20 CV fundamental tasks and achieves SOTA results on 28 public test sets.
VIMER-StrucTexT
End-to-end document O