assert self.alpha_lookup_idx is None or self.alpha_lookup_idx[start_node.name] is None

[AvrahamRaviv]

I got an assert in relu.py, line 227:
assert self.alpha_lookup_idx is None or self.alpha_lookup_idx[start_node.name] is None
It already mentioned here:
#26 (comment)

Here my ONNX graph:

There is no 2 consecutive linear layers, as described in early issue, so what can be cause to this assert?
Thanks!

[AvrahamRaviv]

@shizhouxing
Here is my yaml file:
model:
name: my_model
path: models/my_model_weights.pt
tau: 0.5
data:
dataset: my_dataset
start: 0
end: 10
std: [1.]
mean: [0.]
test_csv: models/dataest_10.csv
num_outputs: 1
specification:
epsilon: 0.001
norm: .inf
robustness_type: SpecificationAllPositive
attack:
pgd_restarts: 2
solver:
batch_size: 1
min_batch_size_ratio: 1
alpha-crown:
disable_optimization: ['MaxPool']
beta-crown:
iteration: 20
lr_beta: 0.03
mip:
parallel_solvers: 8
solver_threads: 4
refine_neuron_time_percentage: 0.8
skip_unsafe: True
bab:
timeout: 180
pruning_in_iteration: False
sort_domain_interval: 1
branching:
method: nonlinear
candidates: 3
nonlinear_split:
num_branches: 2
method: shortcut
filter: true

[shizhouxing]

@AvrahamRaviv There is some bug related to conv for models with complex architectures. Could you try if setting --conv_mode matrix fixes the issue?

[AvrahamRaviv]

@shizhouxing

Will try soon, thanks!