### Install TorchLens and Dependencies

Source: https://github.com/johnmarktaylor91/torchlens/blob/main/README.md

This snippet provides the bash commands to install TorchLens and its required dependency, Graphviz. Graphviz is necessary for generating network visualizations.

```Bash
sudo apt install graphviz
pip install torchlens
```

--------------------------------

### Retrieve Model Metadata

Source: https://github.com/johnmarktaylor91/torchlens/blob/main/README.md

Get metadata for a PyTorch model without saving any activations. This function is equivalent to calling `log_forward_pass` with `layers_to_save=None` and is useful for understanding the model's structure before deciding which layers to inspect.

```Python
tl.get_model_metadata(model)
```

--------------------------------

### Log Forward Pass with TorchLens

Source: https://github.com/johnmarktaylor91/torchlens/blob/main/README.md

Demonstrates how to use `torchlens.log_forward_pass` to capture intermediate activations and metadata from a PyTorch model (AlexNet) during a forward pass. It saves all layers and uses an unrolled visualization option.

```Python
import torch
import torchvision
import torchlens as tl

alexnet = torchvision.models.alexnet()
x = torch.rand(1, 3, 224, 224)
model_history = tl.log_forward_pass(alexnet, x, layers_to_save='all', vis_opt='unrolled')
print(model_history)
```

--------------------------------

### Log Forward Pass and Visualize PyTorch Model

Source: https://github.com/johnmarktaylor91/torchlens/blob/main/README.md

This snippet demonstrates how to use TorchLens to log the forward pass of a PyTorch model and generate a visualization. It shows the extraction of activations from intermediate layers, specifically the second pass of the first linear layer in a simple recurrent model.

```Python
class SimpleRecurrent(nn.Module):
    def __init__(self):
        super().__init__()
        self.fc = nn.Linear(in_features=5, out_features=5)

    def forward(self, x):
        for r in range(4):
            x = self.fc(x)
            x = x + 1
            x = x * 2
        return x


simple_recurrent = SimpleRecurrent()
model_history = tl.log_forward_pass(simple_recurrent, x,
                                    layers_to_save='all',
                                    vis_opt='rolled')
print(model_history['linear_1_1:2'].tensor_contents)  # second pass of first linear layer

```

--------------------------------

### Access Function Call Stack

Source: https://github.com/johnmarktaylor91/torchlens/blob/main/README.md

Retrieve the function call stack information for a specific layer's output. This provides insights into the execution context and the exact code path that generated the activation.

```Python
print(model_history['conv2d_3'].func_call_stack[8])
```

--------------------------------

### Visualize Model Graph

Source: https://github.com/johnmarktaylor91/torchlens/blob/main/README.md

Generate a visualization of the PyTorch model's computational graph. This function does not save any activations and is purely for architectural understanding.

```Python
tl.show_model_graph(model)
```

--------------------------------

### Validate Model Activations

Source: https://github.com/johnmarktaylor91/torchlens/blob/main/README.md

Perform a validation procedure to ensure the correctness of saved model activations. This involves multiple forward passes and checks against ground-truth outputs and random data.

```Python
tl.validate_model_activations(model, activations)
```

--------------------------------

### Access Layer Information from ModelHistory

Source: https://github.com/johnmarktaylor91/torchlens/blob/main/README.md

Shows how to retrieve specific layer information, such as activations and parameters, from a `ModelHistory` object. It illustrates accessing a layer by its name ('conv2d_3_7'), module name ('features.6'), or ordinal position.

```Python
print(model_history['conv2d_3_7'])  # pulling out layer by its name 
# The following commented lines pull out the same layer:
# model_history['conv2d_3'] you can omit the second number (since strictly speaking it's redundant)
# model_history['conv2d_3_7:1'] colon indicates the pass of a layer (here just one)
# model_history['features.6'] can grab a layer by the module for which it is an output
# model_history[7] the 7th layer overall
# model_history[-17] the 17th-to-last layer
```

--------------------------------

### Log Forward Pass with Specific Layers

Source: https://github.com/johnmarktaylor91/torchlens/blob/main/README.md

Log the forward pass of a PyTorch model, selectively saving activations for specified layers. This allows for memory optimization by avoiding the storage of all intermediate activations. Layers can be specified by name, index, or substring.

```Python
model_history = tl.log_forward_pass(alexnet, x, vis_opt='unrolled',
                                    layers_to_save=['conv2d_3_7', 'features', -5, 'linear'])
print(model_history.layer_labels)
```

--------------------------------

### Extract Tensor Activations from Layer

Source: https://github.com/johnmarktaylor91/torchlens/blob/main/README.md

Access the raw tensor contents of a specific layer's output from a model's forward pass history. This is useful for analyzing intermediate results within a neural network.

```Python
print(model_history['conv2d_3_7'].tensor_contents)
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.