Hope to support dynamically loaded libraries ggml-base.so

[yigekuyou]

ggml has support for dynamic loading of backends
but now llama-cpp-python cant support it
This allows you to build multiple backends simultaneously without conflict.

[onestardao]

Yesss this!

Dynamic loading is such a low-key lifesaver — the lack of it has broken more multi-agent setups than I’d care to admit 😅
We've seen this become a blocker in sandboxed RAG deployments too, esp. when trying to mix isolated backends or hot-swap models mid-run.

If you’re mapping out where these frictions accumulate, we’ve actually been building a “failure mode atlas” for exactly this kind of infra issue. Ping me if curious.

Appreciate you surfacing this!