YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

PyTorch mobile ISINSTANCE type OOB PoC

This repository demonstrates an out-of-bounds TypePtr read in PyTorch's lite-interpreter bytecode runtime.

What this PoC shows:

  • a valid .ptl model can be mutated by changing only one serialized ISINSTANCE operand
  • the function's serialized type table still has exactly 1 entry
  • changing ISINSTANCE.N from 1 to 2 makes forward() crash the official torch-2.13.0+cpu wheel with SIGSEGV
  • the malformed model still returns normally on matching list input, but crashes on common non-matching inputs such as a tensor

What this PoC does not show:

  • a stable memory disclosure primitive
  • code execution
  • a claim about every TorchScript loading path

Verified runtime:

  • official torch-2.13.0+cpu wheel

Run:

python3 poc_mobile_isinstance_type_oob.py --isinstance-n 1
python3 poc_mobile_isinstance_type_oob.py --isinstance-n 2
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support