deepseek-ai/DeepSpec
File Complexity*
The distribution of files by complexity: the sum of the McCabe indexes of their units.
File Complexity Overall
  • There are 53 files with units, with 9,564 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)
    • 5 medium complex files (McCabe index sum 51-125; 1,997 lines of code)
    • 8 slightly complex files (McCabe index sum 26-50; 2,482 lines of code)
    • 40 simple files (McCabe index sum 1-25; 5,085 lines of code)
  • 18 main files have no units (e.g. languages without unit analysis, or files without functions) and are not classified.
system0% | 0% | 20% | 25% | 53%
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% | 20% | 25% | 53%
File Complexity per Logical Decomposition
primary
251+
126-250
51-125
26-50
1-25
deepspec0% | 0% | 18% | 25% | 56%
scripts0% | 0% | 36% | 63% | 0%
eval_datasets0% | 0% | 100% | 0% | 0%
config0% | 0% | 0% | 0% | 100%
ROOT0% | 0% | 0% | 0% | 100%
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
target_cache_dataset.py
in deepspec/data
 778 52 106 14 7
generate_train_data.py
in scripts/data
 309 14 78 25 20
confidence_head.py
in deepspec/eval/dspark
 545 22 76 14 14
254 20 59 23 7
config.py
in deepspec/utils
 111 11 55 50 10
base_evaluator.py
in deepspec/eval
 643 22 39 6 6
download_and_split.py
in scripts/data
 164 12 38 23 8
metrics.py
in deepspec/utils
 137 10 36 26 8
jsonl_dataset.py
in deepspec/data
 117 11 32 27 12
base_trainer.py
in deepspec/trainer
 364 18 32 9 7
365 9 30 8 13
optim.py
in deepspec/utils
 124 12 28 23 4
modeling.py
in deepspec/modeling/dspark/gemma4
 568 14 27 5 10
modeling.py
in deepspec/modeling/dspark/qwen3
 494 14 24 5 10
modeling.py
in deepspec/modeling/eagle3/gemma4
 458 15 22 5 4
evaluator.py
in deepspec/eval/dspark
 200 12 22 11 6
distributed.py
in deepspec/utils
 106 12 21 20 4
markov_head.py
in deepspec/modeling/dspark
 242 16 20 8 5
common.py
in deepspec/modeling/eagle3
 165 11 19 12 4
ckpt_manager.py
in deepspec/trainer
 206 10 19 9 4
common.py
in deepspec/modeling/dspark
 257 12 18 7 5
modeling.py
in deepspec/modeling/eagle3/qwen3
 375 14 17 5 4
loss.py
in deepspec/modeling/eagle3
 379 12 17 4 6
parser.py
in deepspec/data
 191 9 16 8 5
hfai_suspend.py
in deepspec/utils
 59 5 14 24 4
cuda_prefetcher.py
in deepspec/data
 43 6 13 30 5
training_logger.py
in deepspec/utils
 82 6 11 13 3
config.py
in deepspec/modeling/dspark/gemma4
 92 3 8 9 5
loss.py
in deepspec/modeling/dspark
 306 8 8 3 1
evaluator.py
in deepspec/eval/eagle3
 163 7 8 5 2
sampling.py
in deepspec/utils
 39 5 7 18 2
eagle3_trainer.py
in deepspec/trainer
 61 4 7 11 4
__init__.py
in deepspec/utils
 73 3 6 8 3
draft_ops.py
in deepspec/eval/dspark
 133 5 5 4 1
io.py
in deepspec/utils
 11 2 4 36 3
config.py
in deepspec/modeling/eagle3/gemma4
 80 3 4 5 2
dspark_trainer.py
in deepspec/trainer
 41 3 3 7 1
train.py
in root
 38 2 3 8 2
eval.py
in root
 59 2 3 5 2
config.py
in deepspec/modeling/eagle3/qwen3
 36 1 1 3 1
config.py
in deepspec/modeling/dspark/qwen3
 52 1 1 2 1
eagle3_qwen3_8b.py
in config/eagle3
 45 1 1 2 1
eagle3_qwen3_14b.py
in config/eagle3
 53 1 1 2 1
eagle3_gemma4_12b.py
in config/eagle3
 53 1 1 2 1
eagle3_qwen3_4b.py
in config/eagle3
 45 1 1 2 1
dspark_gemma4_12b.py
in config/dspark
 61 1 1 2 1
dspark_qwen3_8b.py
in config/dspark
 53 1 1 2 1
dspark_qwen3_14b.py
in config/dspark
 61 1 1 2 1
dspark_qwen3_4b.py
in config/dspark
 53 1 1 2 1
dflash_qwen3_14b.py
in config/dflash
 59 1 1 2 1
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...