yureii1996/cek-probe-model
Untuk cek model apakah maskingan atau bukan
What it solves
It provides a way to verify if an AI model endpoint is actually serving the model it claims to be. It helps users identify "architecture mismatches" where a model's label (e.g., DeepSeek) differs from the actual model responding (e.g., GPT).
How it works
The tool uses Python scripts to send simple identity and logic probes (inference requests) to OpenAI-compatible API endpoints. It compares the model's self-reported identity in the response against the expected model family to detect discrepancies.
Who it’s for
Developers and users of AI model providers who want to ensure the authenticity and correctness of the model endpoints they are using.
Highlights
- Simple Python-based probing scripts.
- Support for multiple testing strategies, including fast probes and scripts with built-in delays/retries for specific providers like InferHub and CFRouter.
- Automatic detection of architecture mismatches based on identity probe responses.
Related
- Project
- Project
- Project
- Project