YOUR MEMORY BUDGET

Local AI models for 12GB VRAM.

Explore models by theoretical weight storage. This shortlist covers available downloadable language models with a known or name-derived parameter count.

10 GBleft for weights at 4-bit
12 GB budget − 2 GB reserve

The reserve is your assumption, not a measured overhead allowance. The default 2 GB may be too little for your runtime or context length. Results are not confirmed fits and have no verified speed rating. Weight estimates use decimal GB and do not include quantization overhead. For timed results on one 12GB machine, see measured speeds.

978 models within the weight budget

Largest parameter count first · not a quality ranking

Mistral AI · Downloadable

Mistral-Nemo-Base-2407

Text generation

Generates text, according to its published model architecture (MistralForCausalLM). No task is listed on Hugging Face, so check the model card for intended use.

Source listed
Jul 18, 2024
Parameters
12.2B
4-bit weights only
≈ 6.1 GB
License
apache-2.0
Compare this model →

Mistral AI · Downloadable

Mistral-Nemo-Instruct-2407

Text generation

Generates text, according to its published model architecture (MistralForCausalLM). No task is listed on Hugging Face, so check the model card for intended use.

Source listed
Jul 17, 2024
Parameters
12.2B
4-bit weights only
≈ 6.1 GB
License
apache-2.0
Compare this model →

OpenBMB · Downloadable

RLAIF-V-12B

Text generation

Generates text for conversations and writing.

Source listed
May 20, 2024
Parameters
12B
4-bit weights only
≈ 6 GB
License
apache-2.0
Compare this model →

Stability AI · Downloadable

stablelm-2-12b

Text generation

Generates text for conversations and writing.

Source listed
Mar 21, 2024
Parameters
12B
4-bit weights only
≈ 6 GB
License
other
Compare this model →

What changes with a 12GB budget?

With 2 GB set aside, 10 GB remains for weights. At 4-bit that allows up to 20 billion parameters in the theoretical calculation. An 8-billion-parameter model alone needs approximately 4 GB at this precision, before its conversation cache and runtime.

Why might a listed model still fail to load?

Its actual weight format may need more space, the runtime may not support its architecture, or the conversation cache may exceed your reserve. Vision models can also need space for image processing. Check the original model card and measure your exact setup before relying on it.

Read the memory guide and technical sources · Compare two models