Change8

b8146

📦 llama-cppView on GitHub →
2 features🐛 10 fixes1 deprecations🔧 6 symbols

Summary

This release introduces significant stability improvements by preventing various integer overflows across ggml and gguf operations, alongside minor fixes and the removal of a deprecated function.

Migration Steps

  1. If you were using `ggml_type_sizef()`, update your code as this function has been removed.

✨ New Features

  • Added overflow tests for GGUF memory size.
  • Added file size check for GGUF arrays.

🐛 Bug Fixes

  • Prevented integer overflow for ggml_context memory size.
  • Fixed integer overflows in `ggml_new_object()`.
  • Prevented string exhaustion in GGUF.
  • Prevented array elements exhaustion in GGUF.
  • Fixed negative tensor type out-of-bounds access.
  • Checked for integer overflow in `ggml_new_tensor_impl` and `ggml_new_object`.
  • Error on duplicate keys when reading GGUF files via Python bindings.
  • Restored `tensor_fields` in Python bindings.
  • Fixed context size calculation when `no_alloc == true` in GGUF.
  • Added printing of values when overflow occurs in ggml.

Affected Symbols

⚡ Deprecations

  • Removed deprecated function `ggml_type_sizef()`.