hello，The fine-tuning of the OtterHD-8B model requires at least how much GPU resources? - Githubissues

Luodian / Otter

🦦 Otter, a multi-modal model based on OpenFlamingo (open-sourced version of DeepMind's Flamingo), trained on MIMIC-IT and showcasing improved instruction-following and in-context learning ability.

https://otter-ntu.github.io/

MIT License

3.52k stars 241 forks source link

hello，The fine-tuning of the OtterHD-8B model requires at least how much GPU resources? #314

Closed dgo2dance closed 7 months ago

dgo2dance commented 7 months ago

hello，The fine-tuning of the OtterHD-8B model requires at least how much GPU resources?

Luodian commented 7 months ago

At least 4xA100, the GPU memory would be the largest bottleneck, you may use 80G GPU to run it. We use 8xA100-80G, and the provided scripts would be easily scaled to larger computes (multiple nodes), as long as you are familiar with slurm.