openai / sparse_autoencoder - Conditional Complexity

Intro

Conditional complexity (also called cyclomatic complexity) is a term used to measure the complexity of software. The term refers to the number of possible paths through a program function. A higher value ofter means higher maintenance and testing costs (infosecinstitute.com).
Conditional complexity is calculated by counting all conditions in the program that can affect the execution path (e.g. if statement, loops, switches, and/or operators, try and catch blocks...).
Conditional complexity is measured at the unit level (methods, functions...).
Units are classified in four categories based on the measured McCabe index: 1-5 (simple units), 6-10 (medium complex units), 11-25 (complex units), 26+ (very complex units).

Learn more...

Conditional Complexity Overall

There are 115 units with 1,128 lines of code in units (41.3% of code).

0 very complex units (0 lines of code)
0 complex units (0 lines of code)
2 medium complex units (106 lines of code)
4 simple units (126 lines of code)
109 very simple units (896 lines of code)

Legend:

51+

26-50

11-25

6-10

1-5

Alternative Visuals

3D view of all units

Conditional Complexity per Extension

51+

26-50

11-25

6-10

1-5

Conditional Complexity per Logical Component

primary logical decomposition

51+

26-50

11-25

6-10

1-5

Most Complex Units

Top 50 most complex units

Unit	# lines	McCabe index	# params
export default function TokenAblationmap() in sae-viewer/src/components/tokenAblationmap.tsx	82	20	3
def predict() in sparse_autoencoder/explanations.py	24	15	2
def batch_parallelize() in sparse_autoencoder/explanations.py	17	7	3
def sharded_grad_norm() in sparse_autoencoder/train.py	18	7	3
export default() in sae-viewer/src/components/featureInfo.tsx	58	6	1
def make_torch_comms() in sparse_autoencoder/train.py	33	6	2
async function fetchData() in sae-viewer/src/components/featureInfo.tsx	30	5	0
def batched() in sparse_autoencoder/explanations.py	13	5	2
def a_star_c_fn() in sparse_autoencoder/explanations.py	5	5	1
def dp_allreduce_() in sparse_autoencoder/train.py	9	5	2
def forward() in sparse_autoencoder/train.py	70	5	2
export default function TokenHeatmap() in sae-viewer/src/components/tokenHeatmap.tsx	28	4	4
def c_fn() in sparse_autoencoder/explanations.py	2	4	1
def init_broadcast_() in sparse_autoencoder/train.py	20	4	2
def sh_allreduce_scale() in sparse_autoencoder/train.py	6	4	2
def _sh_comm_op() in sparse_autoencoder/train.py	10	4	3
function handleClickOutside() in sae-viewer/src/components/tooltip.tsx	3	3	1
export default function FeatureSelect() in sae-viewer/src/components/featureSelect.tsx	37	3	4
export async function get_feature_info() in sae-viewer/src/interpAPI.ts	11	3	2
def get_minimal_subsequence() in sparse_autoencoder/explanations.py	20	3	1
def zero_fn() in sparse_autoencoder/explanations.py	2	3	1
def normalized_mse() in sparse_autoencoder/kernels.py	10	3	2
def all_broadcast() in sparse_autoencoder/train.py	20	3	2
def sharded_topk() in sparse_autoencoder/train.py	22	3	4
def maybe_transpose() in sparse_autoencoder/train.py	2	3	1
def __init__() in sparse_autoencoder/train.py	7	3	2
def logkv() in sparse_autoencoder/train.py	4	3	3
def temporary_weight_swap() in sparse_autoencoder/train.py	8	3	2
function zip_sequence() in sae-viewer/src/components/tokenHeatmap.tsx	8	2	1
export default function Feed() in sae-viewer/src/feed.tsx	28	2	0
export function zip_sequence() in sae-viewer/src/types.ts	8	2	1
export function pathForFeature() in sae-viewer/src/autoencoder_registry.tsx	9	2	1
def predict_many() in sparse_autoencoder/explanations.py	2	2	2
def apply_batched() in sparse_autoencoder/explanations.py	5	2	3
def create_n2g_explanation() in sparse_autoencoder/explanations.py	3	2	0
def first_position_fn() in sparse_autoencoder/explanations.py	2	2	1
def sh_allreduce_forward() in sparse_autoencoder/train.py	13	2	2
def sh_allreduce_backward() in sparse_autoencoder/train.py	14	2	2
def print0() in sparse_autoencoder/train.py	3	2	2
def dumpkvs() in sparse_autoencoder/train.py	4	2	1
def __init__() in sparse_autoencoder/train.py	5	2	3
def main() in sparse_autoencoder/train.py	56	2	0
def v5_32k() in sparse_autoencoder/paths.py	4	2	2
def v5_128k() in sparse_autoencoder/paths.py	4	2	2
def preprocess() in sparse_autoencoder/model.py	5	2	2
def decode() in sparse_autoencoder/model.py	6	2	4
def state_dict() in sparse_autoencoder/model.py	6	2	4
function useOutsideClickAlerter() in sae-viewer/src/components/tooltip.tsx	8	1	2
normalized_activation: scaler() in sae-viewer/src/types.ts	6	1	1
export function interpolateColor() in sae-viewer/src/types.ts	8	1	3