Hugging Face makes a push for open-source custom models

Written By Luka Anicin

Recently, Hugging Face released their version of GPTs named Assistants.

Unlike GPTs, where you need a premium subscription to use them, Assistants are free to create and use.

All created instances are public by default, with the creator's name attached.

A pretty cool option is the capability to try your Assistant with 6 different open-source models, allowing users to find the best fit for their needs.

The list of available models consists of:

  • Mixtral-8x7B

  • Llama2-70B

  • Nous Hermes 2 Mixtral

  • Code Llama 70B

  • Mistral 7B

  • Openchat 3.5

Of course, since the project is new, it could be better. A clear instance of GPT’s superiority in comparison is the ability to browse the web.

Hugging Face is aware of this and has outlined several enhancements they aim to roll out, including integrating the Retrieval-Augmented Generation (RAG) model, parameter modifications, and editing Assistants through API, to name a few.

Nonetheless, this is an excellent opportunity for those curious about customizing but unwilling/unable to sign up for GPT Premium.

Check out the ones already made or dive deep into creating your own Assistant -> https://huggingface.co/chat/assistants

Luka Anicin
Previous

Nvidia gives users custom LLMs
Next

Meta breaks the ice towards transparency