Open magicwang1111 opened 1 year ago
model: arch: mini_gpt4 model_type: pretrain_vicuna freeze_vit: True freeze_qformer: True max_txt_len: 160 end_sym: "###" prompt_path: "prompts/alignment.txt" prompt_template: '###Human: {} ###Assistant: ' ckpt: '/path/to/stage1/checkpoint/'
Is the ckpt setting here pointing to vicuna or pretrained_minigpt4_13B.pth,I have trained 5000 pictures but only 45mb size , I see others only trained 3000 pictures but have 6g size.
ckpt here points to the linear mapping layer (pretrained_minigpt4_13B.pth).
model: arch: mini_gpt4 model_type: pretrain_vicuna freeze_vit: True freeze_qformer: True max_txt_len: 160 end_sym: "###" prompt_path: "prompts/alignment.txt" prompt_template: '###Human: {} ###Assistant: ' ckpt: '/path/to/stage1/checkpoint/'
Is the ckpt setting here pointing to vicuna or pretrained_minigpt4_13B.pth,I have trained 5000 pictures but only 45mb size , I see others only trained 3000 pictures but have 6g size.