How long does it take to finetune or train the models?

THUDM / CogVLM

a state-of-the-art-level open visual language model | 多模态预训练模型

Apache License 2.0

6.07k stars 413 forks source link

How long does it take to finetune or train the models? #398

Closed kynthesis closed 7 months ago

kynthesis commented 8 months ago

Dear CogVLM's authors,

Thank you for your outstanding work on MLLM. Can you share a bit about estimating the time required to fine-tune or train the model?

Hardware requirement

Model Inference:

For INT4 quantization: 1 * RTX 3090(24G) (CogAgent takes ~ 12.6GB, CogVLM takes ~ 11GB)

For FP16: 1 * A100(80G) or 2 * RTX 3090(24G)

Finetuning:

For FP16: 4 * A100(80G) [Recommend] or 8* RTX 3090(24G).

zRzRzRzRzRzRzR commented 7 months ago

what is your means?

Jayantverma2 commented 7 months ago

if i use 4*a100 (80 GB)