Add tracing, factor out some helper functions, fix a few bugs with Nodejs mode #68
Conversation
|
The latest updates on your projects. Learn more about Vercel for Git ↗︎
|
ankrgyl
changed the title
| } | ||
|
|
||
| // https://github.com/openai/openai-node/blob/07b3504e1c40fd929f4aae1651b83afc19e3baf8/src/resources/chat/completions.ts#L43-L49 | ||
| // Updated for https://github.com/openai/openai-node/commit/f10c757d831d90407ba47b4659d9cd34b1a35b1d |
There was a problem hiding this comment.
Is this comment superseded by the next line?
There was a problem hiding this comment.
Not sure. It's just copy/pasted from Vercel's code. Would rather not change it.
packages/proxy/src/proxy.ts
Outdated
| @@ -65,6 +71,12 @@ export interface CacheKeyOptions { | |||
| excludeOrgName?: boolean; | |||
| } | |||
|
|
|||
| export interface SpanLogging { | |||
There was a problem hiding this comment.
nit: for uniformity, should we name this type SpanLogger?
| setName: (name: string) => void; | ||
| log: (args: ExperimentLogPartialArgs) => void; | ||
| end: () => void; | ||
| } |
There was a problem hiding this comment.
do we need to explicitly flush the logger here, or is it guaranteed that the beforeExit event will fire in all environments that we run this?
There was a problem hiding this comment.
I think it's ok to make it the responsibility of the caller (which is currently our own tightly controlled server, where asyncFlush is false by default anyway).
This change is a bunch of small tweaks. To summarize:
fetchdecompresses data, while workerd's does not. Also, we weren't properly closing streams based on how we use Vercel's AI SDK parser, so I just rewrote that to use our own streaming logic and pulled out the parser code.