无法正确load配置文件；Can't load config file correctly

[CallMeFrozenBanana]

你好，我根据readme提供的路径下载了APE-D模型与配置文件
运行脚本如下：
python demo/demo_lazy.py \ --config-file configs/LVISCOCOCOCOSTUFF_O365_OID_VGR_SA1B_REFCOCO_GQA_PhraseCut_Flickr30k/ape_deta/ape_deta_vitl_eva02_clip_vlf_lsj1024_cp_16x4_1080k.py \ --input "./inputs/imgs_0083_339B1005-33B9-4e3c-B4A8-2D7EC23720C4_raw.jpg" \ --output "./seg_results" \ --confidence-threshold 0.1 \ --text-prompt 'car' \ --with-box \ --with-mask \ --with-sseg \ --opts \ train.init_checkpoint="./checkpoints/APE-D/model_final.pth" \ model.model_language.cache_dir="" \ model.model_vision.select_box_nums_for_evaluation=500 \ model.model_vision.text_feature_bank_reset=True \ model.model_vision.backbone.net.xattn=False \ model.model_vision.transformer.encoder.pytorch_attn=True \ model.model_vision.transformer.decoder.pytorch_attn=True \
但获得了错误的结果，以为是config-file错误导致

[CallMeFrozenBanana]

如果换用自带的config文件：configs/LVISCOCOCOCOSTUFF_O365_OID_VGR_SA1B_REFCOCO_GQA_PhraseCut_Flickr30k/ape_deta/ape_deta_vitl_eva02_clip_vlf_lsj1024_cp_16x4_1080k_mdl.py
则是如下报错：
Traceback (most recent call last): File "/home/yenianjin/data/project/APE/demo/demo_lazy.py", line 137, in <module> demo = VisualizationDemo(cfg, args=args) File "/home/yenianjin/data/project/APE/demo/predictor_lazy.py", line 139, in __init__ "__unused_" + "_".join([d for d in cfg.dataloader.train.dataset.names]) File "/home/yenianjin/anaconda3/envs/ape/lib/python3.9/site-packages/omegaconf/listconfig.py", line 176, in __getattr__ self._format_and_raise( File "/home/yenianjin/anaconda3/envs/ape/lib/python3.9/site-packages/omegaconf/base.py", line 190, in _format_and_raise format_and_raise( File "/home/yenianjin/anaconda3/envs/ape/lib/python3.9/site-packages/omegaconf/_utils.py", line 821, in format_and_raise _raise(ex, cause) File "/home/yenianjin/anaconda3/envs/ape/lib/python3.9/site-packages/omegaconf/_utils.py", line 719, in _raise raise ex.with_traceback(sys.exc_info()[2]) # set end OC_CAUSE=1 for full backtrace omegaconf.errors.ConfigAttributeError: ListConfig does not support attribute access full_key: dataloader.train[dataset] object_type=list

[shenyunhang]

你好，我运行了以下命令，并未报错，可能是因为最后一个\要去掉。如果还是有错误，可以把错误截图贴上来看看。

python3.9 demo/demo_lazy.py \
--config-file configs/LVISCOCOCOCOSTUFF_O365_OID_VGR_SA1B_REFCOCO_GQA_PhraseCut_Flickr30k/ape_deta/ape_deta_vitl_eva02_clip_vlf_lsj1024_cp_16x4_1080k.py \
--input "./inputs/imgs_0083_339B1005-33B9-4e3c-B4A8-2D7EC23720C4_raw.jpg" \
--output "./seg_results" \
--confidence-threshold 0.1 \
--text-prompt 'car' \
--with-box \
--with-mask \
--with-sseg \
--opts \
train.init_checkpoint="./checkpoints/APE-D/model_final.pth" \
model.model_language.cache_dir="" \
model.model_vision.select_box_nums_for_evaluation=500 \
model.model_vision.text_feature_bank_reset=True \
model.model_vision.backbone.net.xattn=False \
model.model_vision.transformer.encoder.pytorch_attn=True \
model.model_vision.transformer.decoder.pytorch_attn=True

下面这个配置文件只能训练时候使用，推理时候是会报错的。

configs/LVISCOCOCOCOSTUFF_O365_OID_VGR_SA1B_REFCOCO_GQA_PhraseCut_Flickr30k/ape_deta/ape_deta_vitl_eva02_clip_vlf_lsj1024_cp_16x4_1080k_mdl.py

[CallMeFrozenBanana]

感谢你的回复！
具体的，
使用第一个脚本时不会报错，会出现大量parameter loading相关的warning，以及错误的分割/检测结果(如满图大量的car或者没有结果)
WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.backbone.net.pos_embed' to the model due to incompatible shapes: (1, 197, 1024) in the checkpoint but (1, 442, 1024) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.backbone.net.rope_win.freqs_cos' to the model due to incompatible shapes: (256, 64) in the checkpoint but (1024, 64) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.backbone.net.rope_win.freqs_sin' to the model due to incompatible shapes: (256, 64) in the checkpoint but (1024, 64) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.backbone.net.blocks.0.attn.rope.freqs_cos' to the model due to incompatible shapes: (256, 64) in the checkpoint but (1024, 64) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.backbone.net.blocks.0.attn.rope.freqs_sin' to the model due to incompatible shapes: (256, 64) in the checkpoint but (1024, 64) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.backbone.net.blocks.1.attn.rope.freqs_cos' to the model due to incompatible shapes: (256, 64) in the checkpoint but (1024, 64) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.backbone.net.blocks.1.attn.rope.freqs_sin' to the model due to incompatible shapes: (256, 64) in the checkpoint but (1024, 64) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.backbone.net.blocks.2.attn.rope.freqs_cos' to the model due to incompatible shapes: (256, 64) in the checkpoint but (4096, 64) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.backbone.net.blocks.2.attn.rope.freqs_sin' to the model due to incompatible shapes: (256, 64) in the checkpoint but (4096, 64) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.backbone.net.blocks.3.attn.rope.freqs_cos' to the model due to incompatible shapes: (256, 64) in the checkpoint but (1024, 64) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.backbone.net.blocks.3.attn.rope.freqs_sin' to the model due to incompatible shapes: (256, 64) in the checkpoint but (1024, 64) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.backbone.net.blocks.4.attn.rope.freqs_cos' to the model due to incompatible shapes: (256, 64) in the checkpoint but (1024, 64) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.backbone.net.blocks.4.attn.rope.freqs_sin' to the model due to incompatible shapes: (256, 64) in the checkpoint but (1024, 64) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.backbone.net.blocks.6.attn.rope.freqs_cos' to the model due to incompatible shapes: (256, 64) in the checkpoint but (1024, 64) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.backbone.net.blocks.6.attn.rope.freqs_sin' to the model due to incompatible shapes: (256, 64) in the checkpoint but (1024, 64) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.backbone.net.blocks.7.attn.rope.freqs_cos' to the model due to incompatible shapes: (256, 64) in the checkpoint but (1024, 64) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.backbone.net.blocks.7.attn.rope.freqs_sin' to the model due to incompatible shapes: (256, 64) in the checkpoint but (1024, 64) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.backbone.net.blocks.8.attn.rope.freqs_cos' to the model due to incompatible shapes: (256, 64) in the checkpoint but (4096, 64) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.backbone.net.blocks.8.attn.rope.freqs_sin' to the model due to incompatible shapes: (256, 64) in the checkpoint but (4096, 64) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.backbone.net.blocks.9.attn.rope.freqs_cos' to the model due to incompatible shapes: (256, 64) in the checkpoint but (1024, 64) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.backbone.net.blocks.9.attn.rope.freqs_sin' to the model due to incompatible shapes: (256, 64) in the checkpoint but (1024, 64) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.backbone.net.blocks.10.attn.rope.freqs_cos' to the model due to incompatible shapes: (256, 64) in the checkpoint but (1024, 64) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.backbone.net.blocks.10.attn.rope.freqs_sin' to the model due to incompatible shapes: (256, 64) in the checkpoint but (1024, 64) in the model! You mig ............... WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.model_language.net.text.transformer.resblocks.3.ln_2.weight' to the model due to incompatible shapes: (768,) in the checkpoint but (1280,) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.model_language.net.text.transformer.resblocks.3.ln_2.bias' to the model due to incompatible shapes: (768,) in the checkpoint but (1280,) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.model_language.net.text.transformer.resblocks.3.mlp.c_fc.weight' to the model due to incompatible shapes: (3072, 768) in the checkpoint but (5120, 1280) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.model_language.net.text.transformer.resblocks.3.mlp.c_fc.bias' to the model due to incompatible shapes: (3072,) in the checkpoint but (5120,) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.model_language.net.text.transformer.resblocks.3.mlp.c_proj.weight' to the model due to incompatible shapes: (768, 3072) in the checkpoint but (1280, 5120) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.model_language.net.text.transformer.resblocks.3.mlp.c_proj.bias' to the model due to incompatible shapes: (768,) in the checkpoint but (1280,) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.model_language.net.text.transformer.resblocks.4.ln_1.weight' to the model due to incompatible shapes: (768,) in the checkpoint but (1280,) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.model_language.net.text.transformer.resblocks.4.ln_1.bias' to the model due to incompatible shapes: (768,) in the checkpoint but (1280,) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.model_language.net.text.transformer.resblocks.4.attn.in_proj_weight' to the model due to incompatible shapes: (2304, 768) in the checkpoint but (3840, 1280) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.model_language.net.text.transformer.resblocks.4.attn.in_proj_bias' to the model due to incompatible shapes: (2304,) in the checkpoint but (3840,) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.model_language.net.text.transformer.resblocks.4.attn.out_proj.weight' to the model due to incompatible shapes: (768, 768) in the checkpoint but (1280, 1280) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.model_language.net.text.transformer.resblocks.4.attn.out_proj.bias' to the model due to incompatible shapes: (768,) in the checkpoint but (1280,) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.model_language.net.text.transformer.resblocks.4.ln_2.weight' to the model due to incompatible shapes: (768,) in the checkpoint but (1280,) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.model_language.net.text.transformer.resblocks.4.ln_2.bias' to the model due to incompatible shapes: (768,) in the checkpoint but (1280,) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.model_language.net.text.transformer.resblocks.4.mlp.c_fc.weight' to the model due to incompatible shapes: (3072, 768) in the checkpoint but (5120, 1280) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.model_language.net.text.transformer.resblocks.4.mlp.c_fc.bias' to the model due to incompatible shapes: (3072,) in the checkpoint but (5120,) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.model_language.net.text.transformer.resblocks.4.mlp.c_proj.weight' to the model due to incompatible shapes: (768, 3072) in the checkpoint but (1280, 5120) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.model_language.net.text.transformer.resblocks.4.mlp.c_proj.bias' to the model due to incompatible shapes: (768,) in the checkpoint but (1280,) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.model_language.net.text.transformer.resblocks.5.ln_1.weight' to the model due to incompatible shapes: (768,) in the checkpoint but (1280,) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.model_language.net.text.transformer.resblocks.5.ln_1.bias' to the model due to incompatible shapes: (768,) in the checkpoint but (1280,) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.model_language.net.text.transformer.resblocks.5.attn.in_proj_weight' to the model due to incompatible shapes: (2304, 768) in the checkpoint but (3840, 1280) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.model_language.net.text.transformer.resblocks.5.attn.in_proj_bias' to the model due to incompatible shapes: (2304,) in the checkpoint but (3840,) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.model_language.net.text.transformer.resblocks.5.attn.out_proj.weight' to the model due to incompatible shapes: (768, 768) in the checkpoint but (1280, 1280) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.model_language.net.text.transformer.resblocks.5.attn.out_proj.bias' to the model due to incompatible shapes: (768,) in the checkpoint but (1280,) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.model_language.net.text.transformer.resblocks.5.ln_2.weight' to the model due to incompatible shapes: (768,) in the checkpoint but (1280,) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.model_language.net.text.transformer.resblocks.5.ln_2.bias' to the model due to incompatible shapes: (768,) in the checkpoint but (1280,) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.model_language.net.text.transformer.resblocks.5.mlp.c_fc.weight' to the model due to incompatible shapes: (3072, 768) in the checkpoint but (5120, 1280) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.model_language.net.text.transformer.resblocks.5.mlp.c_fc.bias' to the model due to incompatible shapes: (3072,) in the checkpoint but (5120,) in the model! You might want to double check if this is expected. WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Skip loading parameter 'model_vision.model_language.net.text.transformer.resblocks.5.mlp.c_proj.weight' to the model due to incompatible shapes: (768, 3072) in the checkpoint but (1280, 5120) in the model! You might want to double check if this is expected. ........................... WARNING [01/17 15:20:01 fvcore.common.checkpoint]: Some model parameters or buffers are not found in the checkpoint: model_vision.backbone.net.blocks.0.attn.inner_attn_ln.{bias, weight} model_vision.backbone.net.blocks.0.attn.rope.{freqs_cos, freqs_sin} model_vision.backbone.net.blocks.1.attn.inner_attn_ln.{bias, weight} model_vision.backbone.net.blocks.1.attn.rope.{freqs_cos, freqs_sin} model_vision.backbone.net.blocks.10.attn.inner_attn_ln.{bias, weight} model_vision.backbone.net.blocks.10.attn.rope.{freqs_cos, freqs_sin} model_vision.backbone.net.blocks.11.attn.inner_attn_ln.{bias, weight} model_vision.backbone.net.blocks.12.attn.inner_attn_ln.{bias, weight} model_vision.backbone.net.blocks.12.attn.rope.{freqs_cos, freqs_sin} model_vision.backbone.net.blocks.13.attn.inner_attn_ln.{bias, weight} model_vision.backbone.net.blocks.13.attn.rope.{freqs_cos, freqs_sin} model_vision.backbone.net.blocks.14.attn.inner_attn_ln.{bias, weight} model_vision.backbone.net.blocks.15.attn.inner_attn_ln.{bias, weight} model_vision.backbone.net.blocks.15.attn.rope.{freqs_cos, freqs_sin} model_vision.backbone.net.blocks.16.attn.inner_attn_ln.{bias, weight} model_vision.backbone.net.blocks.16.attn.rope.{freqs_cos, freqs_sin} model_vision.backbone.net.blocks.17.attn.inner_attn_ln.{bias, weight} model_vision.backbone.net.blocks.18.attn.inner_attn_ln.{bias, weight} model_vision.backbone.net.blocks.18.attn.rope.{freqs_cos, freqs_sin} model_vision.backbone.net.blocks.19.attn.inner_attn_ln.{bias, weight} model_vision.backbone.net.blocks.19.attn.rope.{freqs_cos, freqs_sin} model_vision.backbone.net.blocks.2.attn.inner_attn_ln.{bias, weight} model_vision.backbone.net.blocks.20.attn.inner_attn_ln.{bias, weight} model_vision.backbone.net.blocks.21.attn.inner_attn_ln.{bias, weight} model_vision.backbone.net.blocks.21.attn.rope.{freqs_cos, freqs_sin} model_vision.backbone.net.blocks.22.attn.inner_attn_ln.{bias, weight} model_vision.backbone.net.blocks.22.attn.rope.{freqs_cos, freqs_sin} model_vision.backbone.net.blocks.23.attn.inner_attn_ln.{bias, weight} model_vision.backbone.net.blocks.3.attn.inner_attn_ln.{bias, weight} model_vision.backbone.net.blocks.3.attn.rope.{freqs_cos, freqs_sin} model_vision.backbone.net.blocks.4.attn.inner_attn_ln.{bias, weight} model_vision.backbone.net.blocks.4.attn.rope.{freqs_cos, freqs_sin} model_vision.backbone.net.blocks.5.attn.inner_attn_ln.{bias, weight} model_vision.backbone.net.blocks.6.attn.inner_attn_ln.{bias, weight} model_vision.backbone.net.blocks.6.attn.rope.{freqs_cos, freqs_sin} model_vision.backbone.net.blocks.7.attn.inner_attn_ln.{bias, weight} model_vision.backbone.net.blocks.7.attn.rope.{freqs_cos, freqs_sin} model_vision.backbone.net.blocks.8.attn.inner_attn_ln.{bias, weight} model_vision.backbone.net.blocks.9.attn.inner_attn_ln.{bias, weight} model_vision.backbone.net.blocks.9.attn.rope.{freqs_cos, freqs_sin} model_vision.backbone.net.pos_embed model_vision.backbone.net.rope_win.{freqs_cos, freqs_sin} model_vision.criterion.0.fed_loss_pad_classes model_vision.criterion.2.fed_loss_cls_weights model_vision.model_language.net.logit_scale model_vision.model_language.net.text.ln_final.{bias, weight} model_vision.model_language.net.text.token_embedding.weight model_vision.model_language.net.text.transformer.resblocks.0.attn.out_proj.{bias, weight} model_vision.model_language.net.text.transformer.resblocks.0.attn.{in_proj_bias, in_proj_weight} model_vision.model_language.net.text.transformer.resblocks.0.ln_1.{bias, weight} ................ model_vision.transformer.encoder.vl_layers.3.b_attn.layer_norm_l.{bias, weight} [93/1923]model_vision.transformer.encoder.vl_layers.3.b_attn.layer_norm_v.{bias, weight} model_vision.transformer.encoder.vl_layers.3.b_attn.{gamma_l, gamma_v} model_vision.transformer.encoder.vl_layers.4.b_attn.attn.l_proj.{bias, weight} model_vision.transformer.encoder.vl_layers.4.b_attn.attn.out_l_proj.{bias, weight} model_vision.transformer.encoder.vl_layers.4.b_attn.attn.out_v_proj.{bias, weight} model_vision.transformer.encoder.vl_layers.4.b_attn.attn.v_proj.{bias, weight} model_vision.transformer.encoder.vl_layers.4.b_attn.attn.values_l_proj.{bias, weight} model_vision.transformer.encoder.vl_layers.4.b_attn.attn.values_v_proj.{bias, weight} model_vision.transformer.encoder.vl_layers.4.b_attn.layer_norm_l.{bias, weight} model_vision.transformer.encoder.vl_layers.4.b_attn.layer_norm_v.{bias, weight} model_vision.transformer.encoder.vl_layers.4.b_attn.{gamma_l, gamma_v} model_vision.transformer.encoder.vl_layers.5.b_attn.attn.l_proj.{bias, weight} model_vision.transformer.encoder.vl_layers.5.b_attn.attn.out_l_proj.{bias, weight} model_vision.transformer.encoder.vl_layers.5.b_attn.attn.out_v_proj.{bias, weight} model_vision.transformer.encoder.vl_layers.5.b_attn.attn.v_proj.{bias, weight} model_vision.transformer.encoder.vl_layers.5.b_attn.attn.values_l_proj.{bias, weight} model_vision.transformer.encoder.vl_layers.5.b_attn.attn.values_v_proj.{bias, weight} model_vision.transformer.encoder.vl_layers.5.b_attn.layer_norm_l.{bias, weight} model_vision.transformer.encoder.vl_layers.5.b_attn.layer_norm_v.{bias, weight} model_vision.transformer.encoder.vl_layers.5.b_attn.{gamma_l, gamma_v} WARNING [01/17 15:20:01 fvcore.common.checkpoint]: The checkpoint state_dict contains keys that are not used by the model: model_language.net.text.{logit_scale, positional_embedding, text_projection} model_language.net.text.transformer.resblocks.0.ln_1.{bias, weight} model_language.net.text.transformer.resblocks.0.attn.{in_proj_bias, in_proj_weight} model_language.net.text.transformer.resblocks.0.attn.out_proj.{bias, weight} model_language.net.text.transformer.resblocks.0.ln_2.{bias, weight} model_language.net.text.transformer.resblocks.0.mlp.c_fc.{bias, weight} model_language.net.text.transformer.resblocks.0.mlp.c_proj.{bias, weight} model_language.net.text.transformer.resblocks.1.ln_1.{bias, weight} model_language.net.text.transformer.resblocks.1.attn.{in_proj_bias, in_proj_weight} model_language.net.text.transformer.resblocks.1.attn.out_proj.{bias, weight} model_language.net.text.transformer.resblocks.1.ln_2.{bias, weight} model_language.net.text.transformer.resblocks.1.mlp.c_fc.{bias, weight} model_language.net.text.transformer.resblocks.1.mlp.c_proj.{bias, weight} model_language.net.text.transformer.resblocks.2.ln_1.{bias, weight} model_language.net.text.transformer.resblocks.2.attn.{in_proj_bias, in_proj_weight} model_language.net.text.transformer.resblocks.2.attn.out_proj.{bias, weight} model_language.net.text.transformer.resblocks.2.ln_2.{bias, weight} model_language.net.text.transformer.resblocks.2.mlp.c_fc.{bias, weight} model_language.net.text.transformer.resblocks.2.mlp.c_proj.{bias, weight} model_language.net.text.transformer.resblocks.3.ln_1.{bias, weight} model_language.net.text.transformer.resblocks.3.attn.{in_proj_bias, in_proj_weight} model_language.net.text.transformer.resblocks.3.attn.out_proj.{bias, weight} model_language.net.text.transformer.resblocks.3.ln_2.{bias, weight} model_language.net.text.transformer.resblocks.3.mlp.c_fc.{bias, weight} model_language.net.text.transformer.resblocks.3.mlp.c_proj.{bias, weight} model_language.net.text.transformer.resblocks.4.ln_1.{bias, weight} model_language.net.text.transformer.resblocks.4.attn.{in_proj_bias, in_proj_weight} model_language.net.text.transformer.resblocks.4.attn.out_proj.{bias, weight} model_language.net.text.transformer.resblocks.4.ln_2.{bias, weight} model_language.net.text.transformer.resblocks.4.mlp.c_fc.{bias, weight} model_language.net.text.transformer.resblocks.4.mlp.c_proj.{bias, weight} model_language.net.text.transformer.resblocks.5.ln_1.{bias, weight} model_language.net.text.transformer.resblocks.5.attn.{in_proj_bias, in_proj_weight} model_language.net.text.transformer.resblocks.5.attn.out_proj.{bias, weight} model_language.net.text.transformer.resblocks.5.ln_2.{bias, weight} model_language.net.text.transformer.resblocks.5.mlp.c_fc.{bias, weight} model_language.net.text.transformer.resblocks.5.mlp.c_proj.{bias, weight} model_language.net.text.transformer.resblocks.6.ln_1.{bias, weight} model_language.net.text.transformer.resblocks.6.attn.{in_proj_bias, in_proj_weight} model_language.net.text.transformer.resblocks.6.attn.out_proj.{bias, weight} ....
warning很长，就没有截取全了

[shenyunhang]

看样子应该是模型的权重文件有问题。
确保下载的权重是APE-D的。
APE不同版本的配置文件和权重是对应的。