Troubleshooting tool_call Errors with Default LLMs

·

hi guys, anyone had "tool_call" errors with default LLMs lately ? one of our workspace's users has a conversation dedicated to one topic, that they use on a regular basis, but hasn't been able to use any LLM for several days : here's a screenshot of the error log, I suppose the issue comes from the conversation : could it be that it is too rich/long and is therefore causing errors ? cheers

6 comments

Remi
·
·
Thanks Hélène Bonnet for flagging, I will share this with the team 🙏
Wendy Zhao
·
·
Hi Hélène Bonnet, our team recently deployed some fixes addressing this, can you retry in a new conversation?
Hélène Bonnet
·
·
hi Wendy Zhao Remi I've tried to "retry" the agents, the issue remains, for gpt4 and Claude alike
Hélène Bonnet
·
·
hey guys, this issue is not limited to one user on our workspace : I've had other complaints about this could it be because the conversation context is too big ? cheers
Remi
·
·
Hi Hélène Bonnet 👋 Yes, it's likely due to the fact that the conversation is too long and exceed the context window limit 😕
Tomas
·
·
Remi isn't there anything in place that trims the context from the top to avoid this? Or summarizes it from the top (from the initial message) to shorten and squeeze into the context window?