- Newest
- Most votes
- Most comments
Hi Matt,
as you have seen in my post before, I am experiencing the same problem.
In the response you posted, too, the queries the LLM formulated are faulty. They look just like the faulty queries my agents have started writing... As you mention, this hasn't been a problem before and just started two days back. Also, GPT-3 has long been able to handle similar tasks with ease. Really hope AWS fixes this ASAP!
To point out the problem even clearer if that isn't immediately obvious from your cited response:
<function_calls>
<function_calls>
<invoke>
<tool_name>GET::x_amz_knowledgebase_VRVUWUVNFV::Search</tool_name>
<parameters>
<$PARAMETER_NAME>searchQuery</$PARAMETER_NAME> // <-- $PARAMER_NAME-placeholders should prob be "searchQuery"
management IP address for device cisco_r1 // <-- this string should prob be inside xml-tags, where "searchQuery" is
</parameters>
</function_calls>
// <-- the "invoke" tag isnt closed
// <-- the (duplicate) first "function_calls" tag isnt closed
Ahh I hadn't bothered looking closely enough. Thanks Georg. The thing is, mine had been working quite well, although I often had to remind the agent to use the knowledge base which I thought was a bit weird, but otherwise everything was good. Just to let you know, when testing in AU region earlier today, I got three "internal server exception" errors in a row when asking Sonnet to search the knowledge base, after which it reverted to as above, again. Fred and Wilma will have something to say about this!
Hi, I'm facing this exact issue with my agent. It adds multiple unnecessary <function_calls> opening tags causing the format error. It happens around 70% of the time and older versions of my agent seem to work ok. Anyone know of a fix?
Relevant content
- asked 8 months ago
