azure / AI-benchmarking-guide

Intro

Unit size measurements show the distribution of size of units of code (methods, functions...).
Units are classified in four categories based on their size (lines of code): 1-20 (small units), 20-50 (medium size units), 51-100 (long units), 101+ (very long units).
You should aim at keeping units small (< 20 lines). Long units may become "bloaters", code that have increased to such gargantuan proportions that they are hard to work with.

Learn more...

Unit Size Overall

There are 109 units with 1,259 lines of code in units (83.8% of code).

0 very long units (0 lines of code)
0 long units (0 lines of code)
16 medium size units (529 lines of code)
22 small units (360 lines of code)
71 very small units (370 lines of code)

Legend:

101+

51-100

21-50

11-20

1-10

Unit Size per Extension

101+

51-100

21-50

11-20

1-10

Unit Size per Logical Component

primary logical decomposition

101+

51-100

21-50

11-20

1-10

Alternative Visuals

3D view of all units

Longest Units

Top 50 longest units

Unit	# lines	McCabe index	# params
def run_model_sizes() in Benchmarks/NVIDIA/GEMMCublasLt.py	48	4	1
def run_benchmark() in Benchmarks/AMD/LLMBenchmark.py	44	10	1
def prepare_datasets() in Benchmarks/NVIDIA/LLMBenchmark.py	42	6	1
def build() in Benchmarks/NVIDIA/HBMBandwidth.py	38	4	1
def build() in Benchmarks/NVIDIA/GEMMCublasLt.py	38	2	1
def save_results() in Benchmarks/NVIDIA/CPUStream.py	33	2	1
def save_results() in Benchmarks/NVIDIA/HBMBandwidth.py	33	2	1
def save_results() in Benchmarks/AMD/HBMBandwidth.py	33	2	1
def run() in Benchmarks/AMD/FIO.py	33	2	1
def run() in Benchmarks/NVIDIA/FIO.py	32	2	1
def build() in Benchmarks/NVIDIA/CPUStream.py	30	3	1
def build() in Benchmarks/NVIDIA/NCCLBandwidth.py	26	3	1
def build() in Benchmarks/AMD/RCCLBandwidth.py	26	8	1
def run() in Benchmarks/AMD/RCCLBandwidth.py	26	6	1
def run() in Benchmarks/NVIDIA/NCCLBandwidth.py	24	5	1
def run_model_sizes() in Benchmarks/AMD/GEMMHipblasLt.py	23	4	1
def run() in Benchmarks/NVIDIA/CPUStream.py	20	3	1
def run() in Benchmarks/NVIDIA/HBMBandwidth.py	20	3	1
def run_benchmark() in Benchmarks/NVIDIA/LLMBenchmark.py	19	4	1
def create_container() in Benchmarks/AMD/LLMBenchmark.py	19	1	1
def create_container() in Benchmarks/AMD/FlashAttention.py	19	1	1
def run() in Benchmarks/AMD/FlashAttention.py	19	2	1
def get_system_specs() in NVIDIA_runner.py	18	3	0
def run() in Benchmarks/NVIDIA/FlashAttention.py	18	2	1
def create_container() in Benchmarks/AMD/GEMMHipblasLt.py	18	1	1
def create_container() in Benchmarks/AMD/RCCLBandwidth.py	18	1	1
def get_system_specs() in AMD_runner.py	17	1	0
def build() in Benchmarks/AMD/GEMMHipblasLt.py	17	3	1
def run() in Benchmarks/AMD/HBMBandwidth.py	17	3	1
def build() in Benchmarks/NVIDIA/Multichase.py	16	2	1
def run() in Benchmarks/NVIDIA/NVBandwidth.py	16	2	1
def build() in Benchmarks/NVIDIA/NVBandwidth.py	15	3	1
def install_requirements() in Benchmarks/NVIDIA/LLMBenchmark.py	14	4	1
def __init__() in Benchmarks/NVIDIA/GEMMCublasLt.py	12	2	6
def parse_json() in Benchmarks/NVIDIA/GEMMCublasLt.py	12	6	3
def parse_json() in Benchmarks/AMD/GEMMHipblasLt.py	12	6	3
def build() in Benchmarks/AMD/TransferBench.py	12	2	1
def run() in Benchmarks/AMD/TransferBench.py	12	2	1
def __init__() in Benchmarks/AMD/GEMMHipblasLt.py	10	1	6
def run_LLMBenchmark() in NVIDIA_runner.py	9	3	0
def build() in Benchmarks/AMD/HBMBandwidth.py	9	2	1
def get_config() in Benchmarks/NVIDIA/CPUStream.py	8	2	2
def get_config() in Benchmarks/NVIDIA/HBMBandwidth.py	8	2	2
def __init__() in Benchmarks/NVIDIA/LLMBenchmark.py	8	1	4
def get_config() in Benchmarks/NVIDIA/LLMBenchmark.py	8	2	2
def run() in Benchmarks/NVIDIA/Multichase.py	8	1	1
def get_config() in Benchmarks/NVIDIA/GEMMCublasLt.py	8	2	2
def get_config() in Benchmarks/NVIDIA/NVBandwidth.py	8	2	2
def get_config() in Benchmarks/NVIDIA/NCCLBandwidth.py	8	2	2
def get_config() in Benchmarks/AMD/GEMMHipblasLt.py	8	2	2