🔥 L40s Server Is Now Live – Just 0.83 credits/hr!
Glows.ai
<-Go Back
How to Download HuggingFace Models on Glows.ai
Tutorial

This tutorial explains how to download HuggingFace models on Glows.ai, using two available methods: Glows.ai Datadrive storage (local download and upload), and instance-based storage (download directly inside an instance).

  • Glows.ai Datadrive storage: Data can be read and written without limits as long as your Storage Space Plan remains valid. Download speed depends on your local network, making this option suitable for users who frequently need to download the same data (e.g., for model serving).
  • Instance-based storage: Data is valid only while the instance is running; once the instance is released, the data will be deleted. Because it shares the bandwidth of the instance’s data center, download speeds are fast. This method is suitable for users who need the data for one-time use (e.g., testing model performance).

Glows.ai Datadrive Storage

This method saves data to Glows.ai Datadrive. Make sure your Space Storage plan provides enough capacity, and allocate that capacity to the DataDrive in the region you intend to use.

Allocate Storage

Suppose you need to download a model of 65GB and plan to use an NVIDIA GeForce RTX 4090 GPU in the TW-03 region. You’ll first need to go to Storage Space and purchase a 100GB storage package.

image-20251125180753151

Then, click the Modify button in the Storage Space interface to allocate 70GB of space to the TW-03 region Datadrive.

image-20251125181002653

Datadrive Client Download

The Data Drive client currently supports downloading models directly from HuggingFace to the corresponding Datadrive of different region. The process works as follows: using your local network, the client downloads HuggingFace model chunks locally, then synchronizes them to the Datadrive.

Instance-based Storage

Create an Instance

This method requires creating an instance on Glows.ai. Suppose you would use an NVIDIA GeForce RTX 4090 GPU in the TW-03 region, with the environment CUDA 12.8 Torch 2.8.0 Base.

image-20251125120458313

Once the instance is created, you can connect to it via SSH or HTTP Port 8888 (JupyterLab).

image-20251125121244598

Download Model Using Commands

JupyterLab is simple to use. The following example demonstrates operations within JupyterLab. Open a new Terminal.

image-20251125121423526

Enter the following command to install HuggingFace’s official model management tool huggingface_hub:

bash
pip install -U huggingface_hub

image-20251125121510970

Once installed, you can use the hf command to download model files directly to the instance. For example, to download openai/gpt-oss-20b into the /gpt-oss-20b directory, use:

bash
hf download openai/gpt-oss-20b --local-dir /gpt-oss-20b

image-20251125122005123

Running HuggingFace Models on Glows.ai

Some frameworks support directly loading and running HuggingFace models, such as Transformers, SGLang, and GPUStack. You can use the software you’re most familiar with for deployment or refer to the tutorials below:

HuggingFace official website also provides usage examples. If you have any questions or suggestions during your implementation on Glows.ai, feel free to contact us through the channels listed in Contact Us.

image-20251126140029352

Contact Us

If you have any questions or suggestions while using Glows.ai, feel free to contact us via email, Discord, or Line.

Email: support@glows.ai

Discord: https://discord.com/invite/glowsai

Line: https://lin.ee/fHcoDgG