infer issue: 'NoneType' object has no attribute 'memory_efficient_attention'

[wdc233]

Namespace(confidence_threshold=0.2, 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=['/6THardDisk/wendongcheng/SemiCD/CDData/WHU-CD-256/A/whucd_00020.png'], opts=['train.init_checkpoint=checkpoints/model_final.pth', 'model.model_vision.select_box_nums_for_evaluation=500', 'model.model_vision.text_feature_bank_reset=True', 'model.model_language.cache_dir='], output='APE_output/whu-cd_pseudo-label_ape_prob/A/', text_prompt='house,building,road,grass,tree,water', video_input=None, webcam=False, with_box=False, with_mask=False, with_sseg=True)
0%| | 0/1 [00:01<?, ?it/s]
Traceback (most recent call last):
File "demo/demo_lazy.py", line 150, in
predictions, visualized_output, visualized_outputs, metadata = demo.run_on_image(
File "/6THardDisk/wendongcheng/DiffMatch/APE/demo/predictor_lazy.py", line 209, in run_on_image
predictions = self.predictor(image, text_prompt, mask_prompt)
File "/6THardDisk/wendongcheng/DiffMatch/APE/ape/engine/defaults.py", line 229, in call
predictions = self.model([inputs])[0]
File "/6THardDisk/wendongcheng/anaconda3-envs/envs/APECD/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1130, in _call_impl
return forward_call(*input, **kwargs)
File "/6THardDisk/wendongcheng/DiffMatch/APE/ape/modeling/ape_deta/ape_deta.py", line 36, in forward
losses = self.model_vision(batched_inputs, do_postprocess=do_postprocess)
File "/6THardDisk/wendongcheng/anaconda3-envs/envs/APECD/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1130, in _call_impl
return forward_call(*input, **kwargs)
File "/6THardDisk/wendongcheng/DiffMatch/APE/ape/modeling/ape_deta/deformable_detr_segm_vl.py", line 372, in forward
features = self.backbone(images.tensor) # output feature dict
File "/6THardDisk/wendongcheng/anaconda3-envs/envs/APECD/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1130, in _call_impl
return forward_call(*input, **kwargs)
File "/6THardDisk/wendongcheng/DiffMatch/APE/ape/modeling/backbone/vit_eva_clip.py", line 883, in forward
bottom_up_features = self.net(x)
File "/6THardDisk/wendongcheng/anaconda3-envs/envs/APECD/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1130, in _call_impl
return forward_call(*input, **kwargs)
File "/6THardDisk/wendongcheng/DiffMatch/APE/ape/modeling/backbone/vit_eva_clip.py", line 751, in forward
x = blk(x)
File "/6THardDisk/wendongcheng/anaconda3-envs/envs/APECD/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1130, in _call_impl
return forward_call(*input, **kwargs)
File "/6THardDisk/wendongcheng/anaconda3-envs/envs/APECD/lib/python3.8/site-packages/fairscale/nn/checkpoint/checkpoint_activations.py", line 171, in _checkpointed_forward
return original_forward(module, *args, **kwargs)
File "/6THardDisk/wendongcheng/DiffMatch/APE/ape/modeling/backbone/vit_eva_clip.py", line 514, in forward
x = self.attn(x, rel_pos_bias=rel_pos_bias, attn_mask=attn_mask)
File "/6THardDisk/wendongcheng/anaconda3-envs/envs/APECD/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1130, in _call_impl
return forward_call(*input, **kwargs)
File "/6THardDisk/wendongcheng/DiffMatch/APE/ape/modeling/backbone/vit_eva_clip.py", line 274, in forward
x = xops.memory_efficient_attention(
AttributeError: 'NoneType' object has no attribute 'memory_efficient_attention'
请问这是由于什么原因呢？

[wdc233]

apex.normalization.FusedLayerNorm not found, will use pytorch implementations
Please 'pip install xformers'
apex.normalization.FusedLayerNorm not found, will use pytorch implementations
======== shape of rope freq torch.Size([1024, 64]) ========
======== shape of rope freq torch.Size([4096, 64]) ========
貌似提示安装xformer，但是一安装就升级torch版本，又和之前别的版本冲突了，压根不行。我想问你用的什么版本呢，我其他都和你一样啊，但到这就不行了

[XuYunqiu]

apex.normalization.FusedLayerNorm not found, will use pytorch implementations Please 'pip install xformers' apex.normalization.FusedLayerNorm not found, will use pytorch implementations ======== shape of rope freq torch.Size([1024, 64]) ======== ======== shape of rope freq torch.Size([4096, 64]) ======== 貌似提示安装xformer，但是一安装就升级torch版本，又和之前别的版本冲突了，压根不行。我想问你用的什么版本呢，我其他都和你一样啊，但到这就不行了

Hi, I just meet the same issue.
May I ask have you resolved this issue?

[wdc233]

apex.normalization.FusedLayerNorm not found, will use pytorch implementations Please 'pip install xformers' apex.normalization.FusedLayerNorm not found, will use pytorch implementations ======== shape of rope freq torch.Size([1024, 64]) ======== ======== shape of rope freq torch.Size([4096, 64]) ======== 貌似提示安装xformer，但是一安装就升级torch版本，又和之前别的版本冲突了，压根不行。我想问你用的什么版本呢，我其他都和你一样啊，但到这就不行了

Hi, I just meet the same issue. May I ask have you resolved this issue?

没有解决，但这个问题是因为xformer包，我看别人是直接使用docker镜像来解决环境问题的，docker pull keyk13/ape_cu118:v1你试试这个镜像是否可以

[XuYunqiu]

没有解决，但这个问题是因为xformer包，我看别人是直接使用docker镜像来解决环境问题的，docker pull keyk13/ape_cu118:v1你试试这个镜像是否可以

谢谢 我把xformer关掉了 就可以跑了
model.model_vision.backbone.net.xattn=False

[wdc233]

没有解决，但这个问题是因为xformer包，我看别人是直接使用docker镜像来解决环境问题的，docker pull keyk13/ape_cu118:v1你试试这个镜像是否可以

谢谢 我把xformer关掉了 就可以跑了 model.model_vision.backbone.net.xattn=False

关闭这个之后对推理性能有影响吗你那儿看着

[XuYunqiu]

关闭这个之后对推理性能有影响吗你那儿看着

benchmark上的指标不是很清楚 就跑几张自己网上找的图 看着还行