| Title | xorbitsai inference 0.15.0 to 1.4.1 Deserialization |
|---|
| Description | The inference tool by xorbitsai is an LLM deployment tool. It's used to load, run, and manage LLMs for inference tasks. In the xinference/thirdparty/cosyvoice/cli/model.py file , there's a CWE - 502 vulnerability in the load method. This vulnerability exists in version v1.x. The torch.load function is used without the weights_only=True parameter, allowing arbitrary code execution if malicious files are loaded. This poses security risks like unauthorized access and data leakage.
More details: https://github.com/xorbitsai/inference/issues/3190 |
|---|
| Source | ⚠️ https://github.com/xorbitsai/inference/issues/3190 |
|---|
| User | ybdesire (UID 83239) |
|---|
| Submission | 04/06/2025 16:22 (1 Year ago) |
|---|
| Moderation | 04/15/2025 03:16 (8 days later) |
|---|
| Status | Accepted |
|---|
| VulDB entry | 304679 [Xorbits Inference up to 1.4.1 model.py load deserialization] |
|---|
| Points | 20 |
|---|