gemma4-31b-w4a8-gptq-dflash2-b16
- Task
- Vision-language (image + text to text)
- Category
- Image
- License
- Apache 2.0
- Commercial use
- Yes
Permissive license: commercial use is allowed; keep the copyright and license notice. - Size
- About 31 billion parameters
- Memory to load
- ~19 GB (estimate)
- Format
- GPTQ
- Runs on
- GPU
- Access
- Open download
- Downloads (last 30 days)
- 0
- Likes
- 0
- Last updated
- October 25, 2026
How to run it
These are 4-bit quantized weights for NVIDIA GPUs; load them with a library that supports GPTQ, such as vLLM or Hugging Face Transformers.
Plan for roughly 19 GB of memory to load it (our estimate from its size and format).
The model card on Hugging Face has the exact usage code.
Where to run it
In the cloud
Needs about 19 GB: a 24 GB GPU such as an RTX 4090 or RTX A5000 is enough.
- RunPod · rent a GPU by the hour (affiliate link)
Related vision-language (image + text to text) models
Data from Hugging Face's public API, last checked October 25, 2026. Downloads are Hugging Face's count for the last 30 days.