Molmo 72B

Molmo 72B needs an estimated 51.6 GB of VRAM at Q4, 90.1 GB at Q8 and 169.1 GB at FP16. These are estimates from the 73.3B parameter count, covering weights plus about 15% and 0.5 GB for a short context.

No. Molmo 72B does not fit on a 24 GB, 16 GB, 12 GB or 8 GB GPU at Q4, Q8 or FP16 (estimate).

Yes, on a large Mac. With macOS letting the GPU use about 75% of unified memory, the smallest common Mac memory size that fits Q4 (estimate) is 96 GB, since Q4 needs an estimated 51.6 GB.

Yes. Molmo 72B has open weights, so you can self-host it for full control of your data.

Run Molmo 72B locally

Only text and images. The listed input modalities for Molmo 72B are text and image.

Molmo 72B is rank 146 on Arena Vision (Arena's own rank) with a score of 1045.

Molmo 72B benchmarks

Matched on model kind and release date with no comparable price, the closest alternatives are Llama-3.2-11B-Vision-Instruct by Meta, Llama-3.2-1B by Meta and Qwen2.5 7B Instruct by Alibaba.

Llama-3.2-11B-Vision-InstructMolmo 72B vs Llama-3.2-11B-Vision-InstructLlama-3.2-1BMolmo 72B vs Llama-3.2-1BQwen2.5 7B InstructMolmo 72B vs Qwen2.5 7B Instruct

Molmo 72B is open weights, released under the Apache 2.0 license.

Yes. Molmo 72B is under the Apache 2.0 license, a permissive license that allows commercial use.

The Molmo 72B weights are available in the Hugging Face repository allenai/Molmo-72B-0924.

Run Molmo 72B locally

Molmo 72B was made by Ai2 and released on September 25, 2024.

More from Ai2

Yes. Molmo 72B is open weights.