Article directory
ChatGPT replaces the ceiling: HuggingFace version of ChatGPT is here, no magic, no waiting, take off~
Without further ado, please go to the link
https://huggingface.co/chat/
and ask Chinese questions directly. If you return to English, you can go directly to the browser to translate and synchronize.
If you have experienced it, let me tell you (bian) how the HuggingFace version of ChatGPT can stand up to the ceiling of ChatGPT.
Let's first take a look at what the company behind it, HuggingFace, does.
Introduction to HuggingFace
HuggingFace is a very well-known open source AI community and platform. It provides rich models and datasets for NLP and transformer models. Some key features are as follows:
- Transformers model: HuggingFace provides a unified access interface for transformer models, allowing direct use of well-known pre-trained models such as bert/roberta/gpt-2. These models can be used for tasks such as text classification, text generation, QA, etc.
- Tokenizers: Provide corresponding tokenization tools for various transformer models, including bert tokenizer, GPT-2 tokenizer, etc.
- datasets: Provides many NLP datasets, such as SQuAD, XNLI, Restaurants, etc. And has a unified download/load interface.
- Model Hub: Concentrates a large number of pre-trained models, including the latest transformer models from bert to GPT-3. Users can directly download these models for transfer learning with one click.
- Transformers library: Provides a unified implementation and interface of the transformer model, including BertModel, GPT2LMHeadModel, RobertaForSequenceClassification and other models.
- Other tools: It also includes evaluation indicators, data enhancement tools, configuration tools, etc., which can cover the whole process of NLP project From soup to nuts.
In short, HuggingFace provides a very complete ecosystem for NLP and AI. Its goal is to lower the threshold for scientific research and practical projects, so that more people can easily use high-performance NLP models and frameworks. It has become one of the most well-known and active open source communities in the field of NLP at present.
HuggingFace is a sacred place for large models, just like Github is for programmers, and it also benchmarks against Colob and AI Studio's own infinite computing power running model.
Introducing HuggingChat
Look at the interface again:
Does it feel familiar? Does it look like ChatGPT~
Take a closer look, ChatGPT is based on the GPT3.5/4 model,
and HuggingChat uses OpenAssistant as its first open source chat model to display and integrate.
OpenAssistant is one of the best open source chat models today, and by integrating it, HuggingChat allows users to experience powerful chat functions.
List description:
| presentation layer | Core Technology |
|---|---|
| ChatGPT | GPT3.5/4 |
| HuggingChat | OpenAssistant |
But HuggingChat's ambitions won't stop there. Its goal is to attract more open source chat models to join its platform, forming an open source and rich chat model ecology. In this ecosystem, researchers and developers can choose to develop their own solutions based on different models; end users can also choose and use different open source chat models according to their needs.
look forward to
Outlook
Nvidia AI scientist Jim Fan praised: "HuggingChat is an open source alternative to ChatGPT", and said that "if it is subsequently developed into an application, Hugging Face has an advantage over OpenAI."
My understanding is that
since OpenAI is not actually Open, Open has now become the main advantage of Hugging Face.
If ChatGPT is the moment of the iPhone, then HuggingChat deserves the title of the moment of Android.