- Category
- Developer Tools
- Rank
- No. 3660Tools index
- Pricing
- Open Source
- Type
- TOOL
- GitHub
- 1 stars
- Date
About
SDCGuard replays recorded vLLM prefill and decode runs to show how much of each request's prompt the proposed KV-transfer checksum actually compared. Its gate command returns a hold when a request reused cached prefix that was checked only once, so a passing checksum does not imply full coverage. It currently supports one pinned vLLM setup on an H100.
Tags
vllmkv-cachechecksumllm-inferencetestinggpu
Tech Stack
Python
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.