llama-gemma / patch_auth_test32.py
Leon4gr45's picture
Upload folder using huggingface_hub
cdda9a5 verified
Raw History Blame Contribute Delete
1.11 kB
with open("gateway.py", "r") as f:
code = f.read()
# Let's restore the fact that the user's authorization header might need to be passed down.
# Oh! Wait! What if we just bypass authentication entirely for the internal request?
# We CAN'T if `llama-swap` is reading the system `API_KEY` and enforcing it.
# AND we can't unset `API_KEY` in `start.sh` because then `gateway.py` won't know the `API_KEY` to authenticate users!
# The easiest solution:
# In `gateway.py`, we must send `Authorization: Bearer {API_KEY}` to `llama-swap`.
# Let's verify `API_KEY` is not empty.
# If `API_KEY` IS empty, then `llama-swap` shouldn't require one.
# So `llama_auth_headers()` doing `return {"Authorization": f"Bearer {API_KEY}"} if API_KEY else {}` should work!
# BUT WAIT.
# Why did `llama_auth_headers()` FAIL to authenticate to `llama-swap` earlier?
# Maybe `llama-swap` expects the header to be `X-API-Key`?
# Let's check `llama.cpp` server API key header. It supports `Authorization: Bearer <key>`.
# Does `llama-swap` support `Authorization: Bearer <key>`? YES!
# Let's manually test `llama-swap` if we can run it.