Implementation:Trailofbits Fickling Identify Pytorch File Format

Knowledge Sources	Fickling
Domains	Security, File_Format, Supply_Chain
Last Updated	2026-02-14 14:00 GMT

Overview

Concrete tool for classifying PyTorch file formats and detecting polyglot conditions provided by the Fickling library.

Description

identify_pytorch_file_format calls find_file_properties internally, then applies format matching rules to produce an ordered list of detected formats. The first element is the most likely format. Multiple elements indicate a polyglot file. An empty list means the file is not recognized as any PyTorch format.

Usage

Use this to determine what format a PyTorch file is in, and to detect polyglot conditions that may indicate supply chain attacks.

Code Reference

Source Location

• Repository: fickling
• File: fickling/polyglot.py
• Lines: L276-333

Signature

def identify_pytorch_file_format(
    file: str,
    print_properties: bool = False,
    print_results: bool = False
) -> list[str]:
    """Identify PyTorch file format(s) ranked by likelihood.

    Args:
        file: Path to the file to identify.
        print_properties: Print property discovery to stdout.
        print_results: Print format identification results.

    Returns:
        Ordered list of format names. Possible values:
        "TorchScript v1.4", "TorchScript v1.3",
        "TorchScript v1.0", "TorchScript v1.1",
        "PyTorch v1.3", "PyTorch v0.1.1",
        "PyTorch v0.1.10", "PyTorch model archive format".
        Empty list if unrecognized.
    """

Import

from fickling.polyglot import identify_pytorch_file_format

I/O Contract

Inputs

Name	Type	Required	Description
file	str	Yes	Path to the PyTorch file to identify
print_properties	bool	No	Print property discovery (default: False)
print_results	bool	No	Print format identification results (default: False)

Outputs

Name	Type	Description
(return)	list[str]	Ordered list of detected format names; first is most likely

Usage Examples

Identify File Format

from fickling.polyglot import identify_pytorch_file_format

formats = identify_pytorch_file_format("model.pt", print_results=True)

if len(formats) == 0:
    print("Not a recognized PyTorch file")
elif len(formats) == 1:
    print(f"Format: {formats[0]}")
else:
    print(f"POLYGLOT DETECTED: {formats}")
    print(f"Primary: {formats[0]}, Also: {formats[1:]}")

Related Pages

Implements Principle

• Principle:Trailofbits_Fickling_PyTorch_Format_Identification

Requires Environment

• Environment:Trailofbits_Fickling_PyTorch