openai/transformer-debugger
File Complexity*
The distribution of files by complexity: the sum of the McCabe indexes of their units.
File Complexity Overall
  • There are 96 files with units, with 22,962 lines of code.
    • 0 very complex files (McCabe index sum 251+; 0 lines of code)
    • 0 complex files (McCabe index sum 126-250; 0 lines of code)
    • 7 medium complex files (McCabe index sum 51-125; 3,720 lines of code)
    • 22 slightly complex files (McCabe index sum 26-50; 9,400 lines of code)
    • 67 simple files (McCabe index sum 1-25; 9,842 lines of code)
  • 112 main files have no units (e.g. languages without unit analysis, or files without functions) and are not classified.
system0% | 0% | 16% | 40% | 42%
Legend:
251+
126-250
51-125
26-50
1-25


explore: grouped by folders | grouped by complexity
File Complexity per Extension
251+
126-250
51-125
26-50
1-25
py0% | 0% | 14% | 47% | 38%
tsx0% | 0% | 38% | 0% | 61%
ts0% | 0% | 0% | 26% | 73%
File Complexity per Logical Decomposition
primary
251+
126-250
51-125
26-50
1-25
neuron_explainer0% | 0% | 14% | 47% | 38%
neuron_viewer0% | 0% | 26% | 8% | 65%
Most Complex Files (Top 50)

The McCabe index of a file grows with its size: per 100 lines tells a large but plain file from a dense one, and most complex unit whether the complexity sits in one unit or is spread over many.

File# lines# unitsMcCabe index
(sum)		per 100
lines		most complex
unit
activations_and_metadata.py
in neuron_explainer/activations/derived_scalars
 405 33 69 17 13
derived_scalar_store.py
in neuron_explainer/activations/derived_scalars
 534 42 68 13 8
transformer.py
in neuron_explainer/models
 522 52 68 13 4
hooks.py
in neuron_explainer/models
 230 39 67 29 6
NodeTable.tsx
in neuron_viewer/src/TransformerDebugger/cards/node_table
 976 15 62 6 17
scalar_deriver.py
in neuron_explainer/activations/derived_scalars
 357 41 54 15 5
attention.py
in neuron_explainer/activations/derived_scalars
 696 41 53 8 12
model_component_registry.py
in neuron_explainer/models
 376 19 50 13 11
reconstituter_class.py
in neuron_explainer/activations/derived_scalars
 634 44 47 7 2
model_context.py
in neuron_explainer/models
 306 37 47 15 4
derived_scalar_computation.py
in neuron_explainer/activation_server
 993 34 47 5 6
types.ts
in neuron_viewer/src
 123 4 45 37 12
indexing.py
in neuron_explainer/activations/derived_scalars
 422 34 44 10 4
postprocessing.py
in neuron_explainer/activations/derived_scalars
 556 33 43 8 6
autoencoder_context.py
in neuron_explainer/models
 218 21 40 18 7
interactive_model.py
in neuron_explainer/activation_server
 1106 33 40 4 3
explainer.py
in neuron_explainer/explanations
 419 17 39 9 10
fast_dataclasses.py
in neuron_explainer/fast_dataclasses
 88 8 39 44 24
autoencoder.py
in neuron_explainer/activations/derived_scalars
 416 34 38 9 3
read_routes.py
in neuron_explainer/activation_server
 375 20 36 10 9
multi_pass_scalar_deriver.py
in neuron_explainer/activations/derived_scalars
 254 30 34 13 5
urlParams.ts
in neuron_viewer/src/TransformerDebugger/utils
 194 16 34 18 5
reconstituted.py
in neuron_explainer/activations/derived_scalars
 345 27 28 8 2
config.py
in neuron_explainer/activations/derived_scalars
 129 8 27 21 13
edge_attribution.py
in neuron_explainer/activations/derived_scalars
 429 24 27 6 3
derived_scalar_types.py
in neuron_explainer/activations/derived_scalars
 827 14 26 3 5
direct_effects.py
in neuron_explainer/activations/derived_scalars
 416 23 26 6 3
locations.py
in neuron_explainer/activations/derived_scalars
 188 16 26 14 8
simulator.py
in neuron_explainer/explanations
 586 18 26 4 3
prompt_builder.py
in neuron_explainer/explanations
 73 5 23 32 15
readRequests.ts
in neuron_viewer/src/requests
 116 4 23 20 17
file_utils.py
in neuron_explainer
 63 5 22 35 11
multi_group.py
in neuron_explainer/activations/derived_scalars
 195 15 20 10 4
activation_records.py
in neuron_explainer/activations
 119 12 20 17 4
calibrated_simulator.py
in neuron_explainer/explanations
 116 13 19 16 4
raw_activations.py
in neuron_explainer/activations/derived_scalars
 306 18 18 6 1
hook_graph.py
in neuron_explainer/activations
 245 16 18 7 2
download_from_hf.py
in neuron_explainer/scripts
 84 4 18 21 12
nodePage.tsx
in neuron_viewer/src
 185 6 18 10 3
explanations.py
in neuron_explainer/explanations
 136 6 17 13 12
dst_helpers.py
in neuron_explainer/activation_server
 75 7 17 23 8
paths.ts
in neuron_viewer/src/requests
 50 4 17 34 13
autoencoder.py
in neuron_explainer/models
 105 15 16 15 2
nodes.tsx
in neuron_viewer/src/TransformerDebugger/utils
 97 9 16 16 3
LogitsDisplay.tsx
in neuron_viewer/src/TransformerDebugger/cards
 287 4 15 5 9
CancelablePromise.ts
in neuron_viewer/src/client/core
 108 7 14 13 7
residual.py
in neuron_explainer/activations/derived_scalars
 237 13 13 5 1
activations.py
in neuron_explainer/activations
 154 11 13 8 2
api_client.py
in neuron_explainer
 111 3 13 12 7
requests_and_responses.py
in neuron_explainer/activation_server
 337 4 13 4 7
About This Analysis
  • File complexity is the sum of the McCabe indexes (conditional complexity) of the units (methods, functions...) in a file.
  • Files are classified in five categories based on that sum: 1-25 (simple files), 26-50 (slightly complex files), 51-125 (medium complex files), 126-250 (complex files), 251+ (very complex files). The default bands are five times the unit complexity bands, as a file is a group of units.
  • Only main code files with units are classified. Files without units (languages without unit analysis, or files without functions) have no complexity to measure and are left out rather than counted as simple.
  • A very complex file holds many decisions, which often means it has too many responsibilities: a candidate to split, and a place where changes need more testing.
Learn more...