Skip to main content

​
Overview

The profiler module provides tools to analyze neural network models and generate detailed reports about their computational and memory requirements.

​
profile_model

Profiles a neural network model and generates a comprehensive report.

​
Parameters

​
model
object
required
The neural network model to profile. Must have a layer_sizes attribute.
​
batch_size
int
default:"1"
Batch size for activation memory estimation.
​
output_dir
str
default:"profiling"
Directory where the profiling report JSON file will be saved.

​
Returns

Returns a tuple containing:
  • report (dict): Profiling report with the following structure:
    • timestamp: ISO 8601 timestamp
    • model: Model class name
    • layer_sizes: List of layer dimensions
    • batch_size: Batch size used
    • total_trainable_parameters: Total parameter count
    • layer_wise_parameters: List of per-layer parameter details
    • parameter_memory_mb: Memory footprint for float32, float16, and int8
    • activation_memory: Activation memory details in bytes and MB
  • report_file (Path): Path to the saved JSON report

​
Example

from profiler import profile_model
from student import NeuralNetwork

model = NeuralNetwork(
    layer_sizes=[784, 128, 64, 10],
    activations=['relu', 'relu', 'softmax']
)

report, output_file = profile_model(
    model=model,
    batch_size=32,
    output_dir='profiling'
)

print(f"Total parameters: {report['total_trainable_parameters']}")
print(f"Memory (float32): {report['parameter_memory_mb']['float32']} MB")
print(f"Report saved to: {output_file}")

​
Helper Functions

​
summary_table

Generates a formatted text summary of a profiling report. Parameters:
  • report (dict): Profiling report from profile_model
Returns: Formatted string with parameter counts and memory estimates

​
run_from_config

Runs profiling from a configuration file. Parameters:
  • config_path (str): Path to Python configuration module
Returns: Tuple of (report, output_file)

​
Configuration Format

When using run_from_config, your config file should define: 
# One of the following to define the model:
LAYER_SIZES = [784, 128, 10]
ACTIVATIONS = ['relu', 'softmax']

# Or:
MODEL_CLASS = NeuralNetwork
MODEL_KWARGS = {'layer_sizes': [784, 128, 10], 'activations': ['relu', 'softmax']}

# Or:
def build_model():
    return NeuralNetwork(...)

# Optional:
PROFILE_BATCH_SIZE = 32
PROFILE_OUTPUT_DIR = 'profiling'

​
CLI Usage

python profiler.py --config config.py

​
Notes

  • The profiler performs a forward pass to accurately measure activation memory
  • Memory estimates are provided for three precision modes: float32, float16, and int8
  • Layer-wise parameter counts help identify model bottlenecks
  • All numeric values in the JSON report are JSON-serializable (numpy types converted)

Build docs developers (and LLMs) love

Get started for freeTalk to us