46.7B
- Released
- 2y ago
- Size
- 69Bparameters
Ai2 instruction-tuned Llama 2 70B model, trained with DPO on the open Tulu 2 data mix.
How it compares
Frequently asked questions
Running locally
By our estimate, Tulu 2 DPO 70B needs about 48.6 GB at Q4, 84.8 GB at Q8 and 159.1 GB at FP16 of VRAM. That covers the weights plus headroom for a short context, and longer contexts need more.
No, by our estimates Tulu 2 DPO 70B does not fit fully on a 24 GB GPU even at Q4. It needs more VRAM or offloading to system memory.
Yes, a Mac with at least 96 GB of unified memory should run Tulu 2 DPO 70B at Q4, which needs an estimated 48.6 GB. This assumes macOS gives the GPU about 75% of unified memory.
Privacy and data
Yes, Tulu 2 DPO 70B has open weights, so you can run it on your own hardware and keep prompts on your own machines.
Capabilities and specs
No, Tulu 2 DPO 70B takes text input only.
Quality and benchmarks
Similar models from other labs include Mixtral 8x7B Instruct v0.1 (Mistral AI), SOLAR 10.7B Instruct v1.0 (Upstage) and Ferret (Apple). They are the closest matches by kind and release date.
Mixtral 8x7B Instruct v0.1Tulu 2 DPO 70B vs Mixtral 8x7B Instruct v0.1SOLAR 10.7B Instruct v1.0Tulu 2 DPO 70B vs SOLAR 10.7B Instruct v1.0FerretTulu 2 DPO 70B vs Ferret
Licensing and access
Tulu 2 DPO 70B has open weights under the AI2 ImpACT Low-Risk License license, so you can download and run it yourself.
The Tulu 2 DPO 70B weights are on Hugging Face at allenai/tulu-2-dpo-70b.
Basics
Tulu 2 DPO 70B was made by Ai2. It was released on December 8, 2023.
Yes, Tulu 2 DPO 70B has open weights, so you can run it on your own hardware without API fees.
Generated from Vertical Tabs data, updated November 3, 2026.
Discussion
No comments yet
Be the first to share your thoughts.