for devs (ie the only part of the bubble that is seeing any significant usage), qwen 3.8 and deepseek are fine for %95 of use cases.
there are a bazillion providers that will sell you tokens for peanuts. mix that in with self hosting where you have the resources to do so (its docker containers and a bit or orchestration not rocket science) and we will just see the market normalise to something closer to web hosting.
...and windows monkey IT departments will just buy whatever microsoft is selling.
that future is not worth 1/1000th of whatever todays AI hype cycle is getting shoved down our throats.
</pulling some shit out of my arse>
services:
llama:
image: ghcr.io/ggml-org/llama.cpp:server-cuda
container_name: llama
env_file:
- .env.llama
ports:
- "8081:8081"
expose:
- '8081'
volumes:
- ./models:/models
networks:
- proxy
deploy:
resources:
reservations:
devices:
- driver: nvidia
count: all
capabilities: [gpu]