DCLM 7B

By our estimate, DCLM 7B needs about 5.3 GB at Q4, 8.9 GB at Q8 and 16.3 GB at FP16 of VRAM. That covers the weights plus headroom for a short context, and longer contexts need more.

Yes, by our estimates DCLM 7B fits on an 8 GB GPU at Q4 (about 5.3 GB) and at higher precision on larger cards.

Yes, a Mac with at least 8 GB of unified memory should run DCLM 7B at Q4, which needs an estimated 5.3 GB. This assumes macOS gives the GPU about 75% of unified memory.

Yes, DCLM 7B has open weights, so you can run it on your own hardware and keep prompts on your own machines.

Run DCLM 7B locally

No, DCLM 7B takes text input only.

Similar models from other labs include Llama-Guard-3-8B (Meta), BGE Multilingual Gemma2 (BAAI) and Gemma 2 2B IT (Google DeepMind). They are the closest matches by kind and release date.

Llama-Guard-3-8BDCLM 7B vs Llama-Guard-3-8BBGE Multilingual Gemma2DCLM 7B vs BGE Multilingual Gemma2Gemma 2 2B ITDCLM 7B vs Gemma 2 2B IT

DCLM 7B has open weights under the apple-ascl license, so you can download and run it yourself.

The DCLM 7B weights are on Hugging Face at apple/DCLM-7B.

Run DCLM 7B locally

DCLM 7B was made by Apple. It was released on July 18, 2024.

More from Apple

Yes, DCLM 7B has open weights, so you can run it on your own hardware without API fees.