llama.cpp Release b11727 Fixes Truncated GGUF Data Handling
Release b11727 of llama.cpp introduces bounds and short-read checks to prevent silent tensor corruption when handling truncated GGUF files. These updates specifically target the split and export tools within the repository to ensure invalid inputs are rejected safely. This bug fix ensures data integrity during model conversion and splitting operations, preventing users from wasting time debugging corrupted models. It improves the overall robustness of the AI infrastructure ecosystem built around GGUF files. The patch adds tensor-data bounds checks and short-read checks in `gguf-split.cpp` and `export-lora.cpp`, accompanied by a new regression test to verify that truncated inputs are correctly rejected.
## BACKGROUND
GGUF is a binary file format optimized for fast loading and saving of machine learning models for inference. Llama.cpp is an open-source software library that performs inference on large language models using the GGML tensor library.