Stream: wasi-nn

Topic: SafeTensors

view this post on Zulip Matthew Tamayo-Rios (Mar 18 2024 at 19:00):

It would be nice to support SafeTensors. It seems to gaining some momentum and helps avoid a lot of the issues with malicious models: https://kaitchup.substack.com/p/safe-fast-and-memory-efficient-loading

Safe, Fast, and Memory Efficient Loading of LLMs with Safetensors
How to convert and run your models with safetensors

view this post on Zulip Matthew Tamayo-Rios (Mar 18 2024 at 19:00):

Hi Justin,

I got pulled onto other AI projects unrelated to running models at the edge so I had to back burner this for the moment. During that time a perfect example of what we're trying to protect against came out:

https://www.darkreading.com/application-security/hugging-face-ai-platform-100-malicious-code-execution-models

Hugging Face AI Platform Riddled With 100 Malicious Code-Execution Models
The finding underscores the growing risk of weaponizing publicly available AI models and the need for better security to combat the looming threat.

view this post on Zulip Ralph (Mar 25 2024 at 14:15):

there's also things like this happening all over now: https://github.com/microsoft/aici

GitHub - microsoft/aici: AICI: Prompts as (Wasm) Programs
AICI: Prompts as (Wasm) Programs. Contribute to microsoft/aici development by creating an account on GitHub.

Last updated: Nov 22 2024 at 17:03 UTC