You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Security PoC — ctransformers GGUF loader input validation

Purpose: huntr Model-File-Format (MFV) vulnerability report PoC. Not a real model.

bad_type.gguf is a 57-byte crafted GGUF (v3) whose single tensor declares an out-of-range type field. The vendored (2023-frozen) ggml/llama.cpp parser inside ctransformers uses this file-controlled type value as an index into the fixed type_traits[GGML_TYPE_COUNT] table with no range check, causing an out-of-bounds read when the file is loaded via AutoModelForCausalLM.from_pretrained(...).

Reproduce:

from ctransformers import AutoModelForCausalLM
AutoModelForCausalLM.from_pretrained("bad_type.gguf", model_type="llama")
# -> OSError: access violation reading 0xFFFFFFFFFFFFFFFF  (OOB read)

Access-restricted (gated) — for vulnerability triage only.

Downloads last month: 4

GGUF

Model size

32 params

Architecture

Hardware compatibility

We're not able to determine the quantization variants.

View all variants

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support