Hi :wave: I’m building an app in Glide and using the Open AI integration to utilize the “answer question about a table” action and I keep getting error messages that say, “Answer question about a table: This model’s maximum context length is 4097 tokens, however you requested 6914 tokens (6659 in your prompt; 255 for the completion). Please reduce your prompt; or completion length.” I’ve tried multiple troubleshooting steps like adjusting the source tables, changing the temperature, adjusting prompt size, adjusting prompt, adding a new API key and nothing seems to work. Also, even if I delete my previous actions, it seems that the tokens aren’t being cleared after an attempt and are added to the tokens for the current run. Whenever I seem to use the other Open AI integration actions, I don’t have this problem. It’s been pretty consistent for over a week now, I think. Any thoughts? [CleanShot 2023-09-12 at 10.57.20@2x] [CleanShot 2023-09-12 at 10.16.48@2x] [CleanShot 2023-09-12 at 10.12.58@2x]

In general that’s saying you need to reduce the overall character count going to their service. Try reducing the number of columns or the content inside each column passed from the source table to get it under the token count. Worse case if you have a big essay for instance, you could ask to summarize the essay column in a new summarized column and then point the source table columns to to this summary.

Hey @Jeremy! Thanks for your suggestion. I tried it but it didn’t work. :frowning: I summarized the column I was looking to use AI on then I deselected the other columns. I got a new error message about the embeddings, which is the same thing this post is about. So I only grabbed 12 rows of data and created a new column. Then I updated the Action to get the new column, but still see the same error message. I’m not processing much data at all so I’m not sure how it could be that I’m hitting limits (I doubled checked my OpenAI account and I’m not hitting any limits there either). [CleanShot 2023-09-12 at 15.55.53@2x]

Is there a chance your question is too long? 6659 is a lot anyway.

Hello, I use OPEN AI action… model: gpt-4o Works fine with 4096 tokens max length. But then when i set the actual context window from OPEN AI docs 128,000 tokens it doesn’t work. even if i set maximum answer length to 124000 to take into account my input 1355 tokens + buffer) Any suggestions? Is it a Glide limitation? If so, im ready to pay 2 updates to get the full use of OPEN AI Capacity… :wink: even if this double billing wouldn’t be fair as im paying also for my open AI account usage… [Screenshot 2024-06-10 at 15.20.13] [Screenshot 2024-06-10 at 15.20.24] [Screenshot 2024-06-10 at 15.12.59]

Isnt your message showing that youre requesting 129,355 tokens which exceeds gpt-4o’s 128,000 tokens? Guess I dont understand your question. Also not 100% sure if the OpenAI model “gpt-4o” actually works yet. Havent seen any posts or documentation on that.

[Screenshot 2024-06-10 at 16.01.05] Thanks for answering. To simplify our conversation. maximum token window gpt-4o (128000) I request an answer length (124000) So there is enough for my system prompt + user prompt (1500+1500 = a total 3000 ) plus a buffer (1000 tokens). I confirm the action works with gpt-4o model selected in Glide, but with only 4096 tokens maximum. It should be 128.000 Any specialist? @grumo :wink: Creator of excellent AI QUIZZ template worth trying… Merci

I’ve never used more than 4096 tokens for all my AI apps including QuizAI. However, the Maximum Length tooltip text should be updated to say 124000 (or Glide should let us know what’s the actual limit.)

Thanks I actually need less then 8000 tokens myself most of the time … But maybe up to 80.000 tokens sometimes. “Write my ebook chapter about xyz” from an audio recording of keywords. Also maybe in the Glide AI action parameters: Set maximum tokens used per call And temperature etc…

Have you tried the Glide AI model? I believe the latest one points to gpt-4o. Also, what sentences/terms did you use in your prompt to make it use the full amount of tokens?

[image] robot GPT-4o, the latest model from OpenAI, is now available Announcements Hey Martin, I actually checked again today, and want to clarify something. [image] Whilst those models can have a maximum token window of 128k tokens, they have a limited output token. For gpt-4-0125-preview, that limit was 4096, so your question is valid. gpt-3.5 has this same limit stated on their website. On Microsoft’s website, gpt-4o is confirmed to have the same limit. Claude 3 as well. [image] [image] [image] Gemini seems to have the most amongst the big models, for now. [imag…

Open AI Integration - "Answer question about table" error message

Ask for Help

ThinhDinh June 16, 2024, 10:39pm 11

Topic		Replies	Views
OpenAI "Ask a question about a table" ceased working Report a Bug	24	1438	April 24, 2023
Open AI Table Question Ask for Help	8	770	May 8, 2023
Select Specific Model for OpenAI API Ask for Help	10	1656	October 20, 2023
OpenAI Error "Answer question about a table: Could not get embeddings \| Too Many Requests" Ask for Help api , openai	17	1377	January 2, 2024
Limit on data the gets sent from Open AI Ask for Help	1	363	November 17, 2023

Open AI Integration - "Answer question about table" error message

Related topics