I built Galahad looking for people to test it on one GPU

[corbenicai]

I have build Galahad. It saves the model’s KV cache to disk so it can reuse it after a restart, instead of processing the same context again. It works with llama.cpp, vLLM and SGLang.
I’d love a few people to try it on their own setup. It’s free for non-commercial use on one GPU, on Linux x86-64.
What works? What breaks? Does it help with your workload? All feedback is welcome.