weiqingy commented on issue #280:
URL: https://github.com/apache/flink-agents/issues/280#issuecomment-6162038618
Thanks @wenjin272! No more code changes are planned for this issue. The
provider PRs and the #912 wiring cover it, and #1247 fixed the enum mismatch
across the Java providers.
A few small follow-ups came up. I'll file them as separate issues, since
none blocks the feature:
- A Java `Map` field becomes an empty object schema on OpenAI, Azure OpenAI
and Anthropic, so the model can only return `{}`.
- An enum with `@JsonProperty` on only some constants lists the Java names,
so the reply does not parse. The guide documents this for now.
- Python accepts a blank `model` when the setup is built, unlike Java.
- A durable retry after a parse failure replays the cached response.
Also, the idea you raised on #1129 to render a flat `RowTypeInfo` as JSON
Schema hasn't been filed yet. Happy to file it or pick it up if you'd like.
The user guide is ready: a new page, Calling LLMs > Structured Output, with
Java and Python examples. It covers how to declare a schema and read the
result, the `AUTO` / `NATIVE` / `PROMPT` strategies, native vs. prompt
behavior, per-provider support, and schema rules. I'll open the PR shortly. I'd
let that PR close this issue and track the follow-ups separately. WDYT?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
