Repository navigation
Increased the Agent max loops count and fixed failing agent health checks - #509
satti-hari-krishna-reddy wants to merge 18 commits into
Conversation
…e response format
… large results and improve caching logic
frikky
left a comment
frikky
left a comment
There was a problem hiding this comment.
Please verify the decision output controller.
I doubt this will fix anything, as it does not change the actual output format we are expecting.
| IncludeUsage: true, | ||
| } | ||
|
|
||
| if chatCompletion.ResponseFormat == nil { |
There was a problem hiding this comment.
Does this work for all models? Do we have a fallback if a model fails here?
Also, an array does not fit into "json" and "json_object". Is this taken into account?
If we return [{"decision": "here"}], then that does NOT fit with the "json" or "json_object" return mechanism, as that is an array, which LLMs don't support.
So it would require {"decisions": [{"decision": "here"}]}
|
|
||
| return "", errors.New(fmt.Sprintf("Request from org %s with model '%s' exceeds the maximum number of tokens. %s", org, currentModel, err)) | ||
| } | ||
| // Commented out the potential fix for now as we don't know what the actual reason that's causing "timeout awaiting response headers" |
There was a problem hiding this comment.
It's the HTTP request.
The context of the HTTP request itself is set to be too short. Or the third party provider kills the request.
AKA Transport timeout requires manual HTTP config, as you are doing here.
satti-hari-krishna-reddy
changed the title
No description provided.