Keep defaults? #2
Conversation
|
BTW, it may also make sense to remove defaults for models parameter, instead of #1. |
|
I don't think there was a particular reason, probably just left over when testing. |
| cjson.encode payload | ||
|
|
||
| headers = { | ||
| "Host": parse_url(@api_base).host |
There was a problem hiding this comment.
Host header is part of HTTP spec, not related to any of the APIs. As this library tries to support many environments/clients, in my experience I've had some issues with Lua http clients not correctly setting host header based on the request URL. For that reason I would not remove it here.
There was a problem hiding this comment.
Firstly, in my humble opinion, it appears that we are trying to solve a problem that does not yet exist.
Having tested the LuaSocket with numerous different services without explicitly setting the Host in the library code, I didn’t encounter any problems. Of course, there is always the possibility someone, someday, might encounter an issue in a different environment.
In such an event, we would still have three options:
- Request a proper fix in the appropriate location (i.e. the custom Lua HTTP client).
- Instead of waiting for it to be fixed, set the header in the user code with solution Support additional headers #4.
- As a last resort, we can always modify the Lua-OpenAI code itself, preferably not prematurely.
However, the final decision is ultimately up to you and I will withdraw the last commit if you insist.
The models come and go, it is the client's responsibility to stay updated. P.S. We could ensure that the model is specified, but what's the point? The server returns an error without needing assistance from the library."
Ref: - https://platform.openai.com/docs/api-reference/making-requests Instead, we should implement the ability to include arbitrary headers
2 participants
This is rather a question.
Is there a reason to unconditionally alter openai defaults?