Large Language Models (LLMs)
Large Language Models (LLMs) are generative models capable of generating sophisticated natural language text and can be used to power conversational systems. On this page, you will find resources and tutorials on how to set up and run LLM training or inference workloads on our systems.
Downloading Models¶
At the time of this writing, The Hugging Face Hub is the most common repository for LLMs.
The huggingface_hub Python package contains a command-line interface (CLI) which can be used to download models. For example, to download the model Zephyr-7b-beta, first install huggingface_hub in a virtual environment, then on a login node run:
Disable hf_xet
We set the variable HF_HUB_DISABLE_XET to avoid using the hf_xet package to download models. This package, meant to make downloading artifacts from the Hugging Face Hub more efficient, currently leads to failures on our systems and should not be used at this time.
For more options, please see our article on the Hugging Face ecosystem.