From b069e526deabf7207906ddfae1eb61551b2f612c Mon Sep 17 00:00:00 2001 From: Jai Patel <61297975+jpatel2005@users.noreply.github.com> Date: Wed, 18 Nov 2026 09:28:47 -0500 Subject: [PATCH] gguf-py : fix tensor offset overflow in GGUFReader (#32495) * gguf-py : fix tensor offset overflow in GGUFReader * gguf-py : cast alignment padding to int --- gguf-py/gguf/gguf_reader.py | 6 ++++-- gguf-py/tests/test_gguf_reader_validation.py | 12 ++++++++++-- 2 files changed, 14 insertions(+), 4 deletions(-) diff --git a/gguf-py/gguf/gguf_reader.py b/gguf-py/gguf/gguf_reader.py index bf3e08380..d8b07ef01 100644 --- a/gguf-py/gguf/gguf_reader.py +++ b/gguf-py/gguf/gguf_reader.py @@ -189,7 +189,7 @@ class GGUFReader: raise ValueError('Invalid alignment: must be a non-zero power of two') padding = offs % self.alignment if padding != 0: - offs += self.alignment - padding + offs += int(self.alignment - padding) self.data_offset = offs self._build_tensors(offs, tensors_fields) @@ -350,7 +350,9 @@ class GGUFReader: np_dims = tuple(reversed(dims.tolist())) block_size, type_size = GGML_QUANT_SIZES[ggml_type] n_bytes = n_elems * type_size // block_size - data_offs = int(start_offs + offset_tensor[0]) + data_offs = start_offs + int(offset_tensor[0]) + if data_offs + n_bytes > self.data.nbytes: + raise ValueError(f'Tensor {tensor_name} data exceeds file size {self.data.nbytes}') item_type: npt.DTypeLike if ggml_type == GGMLQuantizationType.F16: item_count = n_elems diff --git a/gguf-py/tests/test_gguf_reader_validation.py b/gguf-py/tests/test_gguf_reader_validation.py index 98f30a969..e6564ed7d 100644 --- a/gguf-py/tests/test_gguf_reader_validation.py +++ b/gguf-py/tests/test_gguf_reader_validation.py @@ -5,7 +5,7 @@ import pytest from gguf.gguf_reader import GGUFReader -def _write_gguf(path, n_dims_field, dims): +def _write_gguf(path, n_dims_field, dims, offset=0): buf = b'GGUF' + struct.pack('