Launching MiniGPT4 after finetuning.

[eduardofarina]

Hi everyone, I've finished the finetuning stage and how can I launch the model API using this last version I've just finetuned? Best,

[Ericsson12resffd0]

понедельник, 1 мая 2023 г. пользователь Eduardo Farina < @.***> написал:

Hi everyone, I've finished the finetuning stage and how can I launch the model API using this last version I've just finetuned? Best,

— Reply to this email directly, view it on GitHub https://github.com/Vision-CAIR/MiniGPT-4/issues/178, or unsubscribe https://github.com/notifications/unsubscribe-auth/A7KUGCGLVR4KKFL7EXI2WKTXD76PXANCNFSM6AAAAAAXSBK5ME . You are receiving this because you are subscribed to this thread.Message ID: @.***>

-- unlocking

[kuoyenlo]

@eduardofarina Do you know how much is the memory requirement for the finetune stage?

[TsuTikgiau]

@eduardofarina hello, you can modify the evaluation checkpoint path to the model checkpoint you fine-tuned in the config file eval_configs/minigpt4_eval.yaml

[eduardofarina]

Thanks @TsuTikgiau !

@kuoyenlo I am using A100 80GB, but for finetuning the 7B llama it required 24GB NVRAM.

[kuoyenlo]

@eduardofarina Thanks for your reply. Did you finetune with the official dataset or your own dataset? Wondering whether the memory depends on the size of dataset.

[TsuTikgiau]

@kuoyenlo Hello! The GPU memory usage depends on the batch size and is not related to the dataset size.