Great work! How many GPU resources used to train the model?

[gxd1994]

No description provided.

[yunkchen]

Our training consists of two stages, and each stage utilized approximately more than 200 A800 GPUs, training for 24*5 hours; based on our pre-trained model, performing fine-tuning with LoRA only requires an A10 or a GPU with equivalent memory performance for 2 hours to achieve satisfactory results.