Skip to content

仅使用eval结果复现的问题 #38

@NLP-1217

Description

@NLP-1217

谢谢up主,特别有意义的工作,其中我进行测试时有点小问题:
利用512和multisiz而进行结果复现,结果发现会有点差距,请问具体的原因是什么?

wukong测试数据为例:
Multisize----SenACC 53.92 NED 7594 FID 27.96 LPIPS 0.1035
512----------SenACC 55.74 NED 75.41 FID 28.53 LPIPS 0.1098

比如:
torchrun --standalone --nnodes=1 --nproc_per_node=2 --master_port $MASTER_PORT
/disks/sdb/user_space/siyuanxue/FluxText/eval/anytext_inference.py
--config_path /disks/sdb/user_space/siyuanxue/FluxText/model_512/config.yaml \ ##这是官网提供的config
--model_path /disks/sdb/user_space/siyuanxue/FluxText/model_512/pytorch_lora_weights.safetensors ##这是官网提供的权重
--json_path /disks/sdb/user_space/siyuanxue/FluxText/AnyText-benchmark/benchmark/laion_word/test1k.json
--output_dir /disks/sdb/user_space/siyuanxue/FluxText/eval/eval_out/laion_word_512

是否释放了最优模型?这部分还请您告知。谢谢!

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions