actnn icon indicating copy to clipboard operation
actnn copied to clipboard
verified publisher icon ucbrise

There is something wrong with loss.backward()

Open Harr7y opened this issue 4 years ago • 2 comments

I just modify the model by

model = actnn.QModule(model)

After that, something wrong happened as follows:

Traceback (most recent call last): File "train.py", line 336, in main() File "train.py", line 332, in main train(args, model) File "train.py", line 212, in train loss.backward() File "/home/hku/anaconda3/envs/torch17/lib/python3.7/site-packages/torch/tensor.py", line 221, in backward torch.autograd.backward(self, gradient, retain_graph, create_graph) File "/home/hku/anaconda3/envs/torch17/lib/python3.7/site-packages/torch/autograd/init.py", line 132, in backward allow_unreachable=True) # allow_unreachable flag RuntimeError: Function linearBackward returned an invalid gradient at index 0 - got [25216, 3072] but expected shape compatible with [128, 197, 3072]

Harr7y avatar Sep 24 '21 13:09 Harr7y

Hi, we cannot know the bug from the backtrace only. Could you provide the architecture of your model? For example, print(model). Some reproducible scripts will be even better.

merrymercy avatar Sep 25 '21 17:09 merrymercy

Hi, we cannot know the bug from the backtrace only. Could you provide the architecture of your model? For example, print(model). Some reproducible scripts will be even better.

Thanks for your reply! Before applying ActNN to it, the model's structure is as follows:

VisionTransformer( (transformer): Transformer( (embeddings): Embeddings( (patch_embeddings): Conv2d(3, 768, kernel_size=(16, 16), stride=(16, 16)) (dropout): Dropout(p=0.1, inplace=False) ) (encoder): Encoder( (layer): ModuleList( (0): Block( (attention_norm): LayerNorm((768,), eps=1e-06, elementwise_affine=True) (ffn_norm): LayerNorm((768,), eps=1e-06, elementwise_affine=True) (ffn): Mlp( (fc1): Linear(in_features=768, out_features=3072, bias=True) (fc2): Linear(in_features=3072, out_features=768, bias=True) (dropout): Dropout(p=0.1, inplace=False) ) (attn): Attention( (query): Linear(in_features=768, out_features=768, bias=True) (key): Linear(in_features=768, out_features=768, bias=True) (value): Linear(in_features=768, out_features=768, bias=True) (out): Linear(in_features=768, out_features=768, bias=True) (attn_dropout): Dropout(p=0.0, inplace=False) (proj_dropout): Dropout(p=0.0, inplace=False) (softmax): Softmax(dim=-1) ) ) (1): Block( (attention_norm): LayerNorm((768,), eps=1e-06, elementwise_affine=True) (ffn_norm): LayerNorm((768,), eps=1e-06, elementwise_affine=True) (ffn): Mlp( (fc1): Linear(in_features=768, out_features=3072, bias=True) (fc2): Linear(in_features=3072, out_features=768, bias=True) (dropout): Dropout(p=0.1, inplace=False) ) (attn): Attention( (query): Linear(in_features=768, out_features=768, bias=True) (key): Linear(in_features=768, out_features=768, bias=True) (value): Linear(in_features=768, out_features=768, bias=True) (out): Linear(in_features=768, out_features=768, bias=True) (attn_dropout): Dropout(p=0.0, inplace=False) (proj_dropout): Dropout(p=0.0, inplace=False) (softmax): Softmax(dim=-1) ) ) (2): Block( (attention_norm): LayerNorm((768,), eps=1e-06, elementwise_affine=True) (ffn_norm): LayerNorm((768,), eps=1e-06, elementwise_affine=True) (ffn): Mlp( (fc1): Linear(in_features=768, out_features=3072, bias=True) (fc2): Linear(in_features=3072, out_features=768, bias=True) (dropout): Dropout(p=0.1, inplace=False) ) (attn): Attention( (query): Linear(in_features=768, out_features=768, bias=True) (key): Linear(in_features=768, out_features=768, bias=True) (value): Linear(in_features=768, out_features=768, bias=True) (out): Linear(in_features=768, out_features=768, bias=True) (attn_dropout): Dropout(p=0.0, inplace=False) (proj_dropout): Dropout(p=0.0, inplace=False) (softmax): Softmax(dim=-1) ) ) (3): Block( (attention_norm): LayerNorm((768,), eps=1e-06, elementwise_affine=True) (ffn_norm): LayerNorm((768,), eps=1e-06, elementwise_affine=True) (ffn): Mlp( (fc1): Linear(in_features=768, out_features=3072, bias=True) (fc2): Linear(in_features=3072, out_features=768, bias=True) (dropout): Dropout(p=0.1, inplace=False) ) (attn): Attention( (query): Linear(in_features=768, out_features=768, bias=True) (key): Linear(in_features=768, out_features=768, bias=True) (value): Linear(in_features=768, out_features=768, bias=True) (out): Linear(in_features=768, out_features=768, bias=True) (attn_dropout): Dropout(p=0.0, inplace=False) (proj_dropout): Dropout(p=0.0, inplace=False) (softmax): Softmax(dim=-1) ) ) (4): Block( (attention_norm): LayerNorm((768,), eps=1e-06, elementwise_affine=True) (ffn_norm): LayerNorm((768,), eps=1e-06, elementwise_affine=True) (ffn): Mlp( (fc1): Linear(in_features=768, out_features=3072, bias=True) (fc2): Linear(in_features=3072, out_features=768, bias=True) (dropout): Dropout(p=0.1, inplace=False) ) (attn): Attention( (query): Linear(in_features=768, out_features=768, bias=True) (key): Linear(in_features=768, out_features=768, bias=True) (value): Linear(in_features=768, out_features=768, bias=True) (out): Linear(in_features=768, out_features=768, bias=True) (attn_dropout): Dropout(p=0.0, inplace=False) (proj_dropout): Dropout(p=0.0, inplace=False) (softmax): Softmax(dim=-1) ) ) (5): Block( (attention_norm): LayerNorm((768,), eps=1e-06, elementwise_affine=True) (ffn_norm): LayerNorm((768,), eps=1e-06, elementwise_affine=True) (ffn): Mlp( (fc1): Linear(in_features=768, out_features=3072, bias=True) (fc2): Linear(in_features=3072, out_features=768, bias=True) (dropout): Dropout(p=0.1, inplace=False) ) (attn): Attention( (query): Linear(in_features=768, out_features=768, bias=True) (key): Linear(in_features=768, out_features=768, bias=True) (value): Linear(in_features=768, out_features=768, bias=True) (out): Linear(in_features=768, out_features=768, bias=True) (attn_dropout): Dropout(p=0.0, inplace=False) (proj_dropout): Dropout(p=0.0, inplace=False) (softmax): Softmax(dim=-1) ) ) (6): Block( (attention_norm): LayerNorm((768,), eps=1e-06, elementwise_affine=True) (ffn_norm): LayerNorm((768,), eps=1e-06, elementwise_affine=True) (ffn): Mlp( (fc1): Linear(in_features=768, out_features=3072, bias=True) (fc2): Linear(in_features=3072, out_features=768, bias=True) (dropout): Dropout(p=0.1, inplace=False) ) (attn): Attention( (query): Linear(in_features=768, out_features=768, bias=True) (key): Linear(in_features=768, out_features=768, bias=True) (value): Linear(in_features=768, out_features=768, bias=True) (out): Linear(in_features=768, out_features=768, bias=True) (attn_dropout): Dropout(p=0.0, inplace=False) (proj_dropout): Dropout(p=0.0, inplace=False) (softmax): Softmax(dim=-1) ) ) (7): Block( (attention_norm): LayerNorm((768,), eps=1e-06, elementwise_affine=True) (ffn_norm): LayerNorm((768,), eps=1e-06, elementwise_affine=True) (ffn): Mlp( (fc1): Linear(in_features=768, out_features=3072, bias=True) (fc2): Linear(in_features=3072, out_features=768, bias=True) (dropout): Dropout(p=0.1, inplace=False) ) (attn): Attention( (query): Linear(in_features=768, out_features=768, bias=True) (key): Linear(in_features=768, out_features=768, bias=True) (value): Linear(in_features=768, out_features=768, bias=True) (out): Linear(in_features=768, out_features=768, bias=True) (attn_dropout): Dropout(p=0.0, inplace=False) (proj_dropout): Dropout(p=0.0, inplace=False) (softmax): Softmax(dim=-1) ) ) (8): Block( (attention_norm): LayerNorm((768,), eps=1e-06, elementwise_affine=True) (ffn_norm): LayerNorm((768,), eps=1e-06, elementwise_affine=True) (ffn): Mlp( (fc1): Linear(in_features=768, out_features=3072, bias=True) (fc2): Linear(in_features=3072, out_features=768, bias=True) (dropout): Dropout(p=0.1, inplace=False) ) (attn): Attention( (query): Linear(in_features=768, out_features=768, bias=True) (key): Linear(in_features=768, out_features=768, bias=True) (value): Linear(in_features=768, out_features=768, bias=True) (out): Linear(in_features=768, out_features=768, bias=True) (attn_dropout): Dropout(p=0.0, inplace=False) (proj_dropout): Dropout(p=0.0, inplace=False) (softmax): Softmax(dim=-1) ) ) (9): Block( (attention_norm): LayerNorm((768,), eps=1e-06, elementwise_affine=True) (ffn_norm): LayerNorm((768,), eps=1e-06, elementwise_affine=True) (ffn): Mlp( (fc1): Linear(in_features=768, out_features=3072, bias=True) (fc2): Linear(in_features=3072, out_features=768, bias=True) (dropout): Dropout(p=0.1, inplace=False) ) (attn): Attention( (query): Linear(in_features=768, out_features=768, bias=True) (key): Linear(in_features=768, out_features=768, bias=True) (value): Linear(in_features=768, out_features=768, bias=True) (out): Linear(in_features=768, out_features=768, bias=True) (attn_dropout): Dropout(p=0.0, inplace=False) (proj_dropout): Dropout(p=0.0, inplace=False) (softmax): Softmax(dim=-1) ) ) (10): Block( (attention_norm): LayerNorm((768,), eps=1e-06, elementwise_affine=True) (ffn_norm): LayerNorm((768,), eps=1e-06, elementwise_affine=True) (ffn): Mlp( (fc1): Linear(in_features=768, out_features=3072, bias=True) (fc2): Linear(in_features=3072, out_features=768, bias=True) (dropout): Dropout(p=0.1, inplace=False) ) (attn): Attention( (query): Linear(in_features=768, out_features=768, bias=True) (key): Linear(in_features=768, out_features=768, bias=True) (value): Linear(in_features=768, out_features=768, bias=True) (out): Linear(in_features=768, out_features=768, bias=True) (attn_dropout): Dropout(p=0.0, inplace=False) (proj_dropout): Dropout(p=0.0, inplace=False) (softmax): Softmax(dim=-1) ) ) (11): Block( (attention_norm): LayerNorm((768,), eps=1e-06, elementwise_affine=True) (ffn_norm): LayerNorm((768,), eps=1e-06, elementwise_affine=True) (ffn): Mlp( (fc1): Linear(in_features=768, out_features=3072, bias=True) (fc2): Linear(in_features=3072, out_features=768, bias=True) (dropout): Dropout(p=0.1, inplace=False) ) (attn): Attention( (query): Linear(in_features=768, out_features=768, bias=True) (key): Linear(in_features=768, out_features=768, bias=True) (value): Linear(in_features=768, out_features=768, bias=True) (out): Linear(in_features=768, out_features=768, bias=True) (attn_dropout): Dropout(p=0.0, inplace=False) (proj_dropout): Dropout(p=0.0, inplace=False) (softmax): Softmax(dim=-1) ) ) ) (encoder_norm): LayerNorm((768,), eps=1e-06, elementwise_affine=True) ) ) (head): Linear(in_features=768, out_features=10, bias=True) )

Harr7y avatar Sep 26 '21 03:09 Harr7y