I ran into issues earlier today when trying to add more data to my custom GPT. My GPT woulnt update and sometime produced an error: ‘Error saving draft when building cusom GPT…’ I did a bit of research and thought it was because I had been feeding it too much data. I thought that the limit was 6-7k words but on the documentation it says 128k whhich is enough for a novel. so whats the deal here? can i create a custom GPT with 128k of custom data or am i limited to 6-7k words? thanks.
Saving problems
“Error saving draft” can simply be the GPT database being temporarily unavailable. If you watch, typing into the input boxes of GPT configure has constant updates in the background into a draft, and some data of this stream not making it over the network might be expected.
It is when you finally “create” that you have to look for signs of failure. For example, I just made one full of trademarked characters and companies, and we see that OpenAI made that failure mode clearer.
File content limitations
First restriction, if your “add more data” is in files:
FAQ:
Q: How many files can I upload to a GPT?
A: We allow up to 20 files to be uploaded to a custom GPT.
What are those file upload size restrictions?
All files uploaded to a GPT or a ChatGPT conversation have a hard limit of 512MB per file.
All text and document files uploaded to a GPT or to a ChatGPT conversation are capped at 2M tokens per file. This limitation does not apply to spreadsheets.
For CSV files or spreadsheets, the file size cannot exceed approximately 50MB, depending on the size of each row.
For images, there’s a limit of 20MB per image.
Additionally, there are usage caps:
Each end-user is capped at 10GB.
Knowing how many tokens OpenAI obtained from a file is opaque.
Instructions
The remaining restriction is the instruction length: GPT instructions cannot be longer than 8000 characters. That cannot be used for extensive data.
3 Likes
Right thanks very much for the reply! So just to clarify, what is the total token limit for data submitted to a custom GPT? So this would be all of the custom data that it has to make it unique.
What makes a GPT unique is just the instructions.
What makes it possibly knowledgeable is uploaded files, but it must have motivation and enhanced information about what it will find to then search. The amount it can pull at once from that kind of data submitted at once is limited and based on a similarity search. That’s like your mini web search.
Code interpreter is a bit harder to sculpt an application around, but that is available, and can have the same files placed within - and exposed directly to the user.
Then you can call upon your external API to fulfill more information or perform tasks that affect the real world.
I am basically using it at the moment to create a dialogue generator for the game I am creating. SO far I havent uploaded any files ive just been crafting out the lore, mythology, culture etc for the NPC’s in my game. I have been doing this purely from within the browser. Every now and again I will ask the GPT to update (which then adds it to its dataset). I can get it to show me the information it has. So this is essentially as far as i can tell a single text file which i am constantly adding to. What i want to know is how large can this file be, ie token limit. To use it I will just be asking it various questions or even getting it to generate questions itself which will be fine tuned based on the infomation I have given it. Thanks again for the reply.
The instruction box (gizmo context) is limited to 8000 characters.
The GPT Builder AI has to resend a complete refinement of the entire instruction each time you ask for an alteration, and won’t approach the approximately 2000 English language tokens that would be.
thanks thats fine but again its not the message length i am concerned with. it is the size of the file which i am progressively building up to fine tune/personalise my GPT.
You seem to be under an illusion that there is something happening to an AI model when you talk to a chatbot about it or something greater is becoming built.
That is incorrect.
The GPT builder AI only emits language directly into the same boxes that you can edit in the configure tab.
“context” translates to “instructions”
Ok ive just looked at the cofigure tab and seen the instructions i have been generating. its this which i would like to know the size limit of, is that 8k tokens? If i put all my context/instructions into a file and uploaded it to the knowledge section could I have a greater instruction set?
This is what I got from askign chatGPT:
The size limit for file uploads can vary depending on the platform or tool you’re using. For OpenAI’s API, which powers GPT-based models, the file upload limit is generally:
- 5 MB per file: The maximum size allowed for any individual file uploaded to OpenAI’s API is 5 megabytes (MB).
If you’re working within a specific platform’s file-uploading features, you may want to check their documentation for more details. For larger sets of information, you can consider splitting your data into multiple files if necessary.
In case you’re handling large amounts of data (like extensive lore and mythology), you might want to prioritize the most important content or break it down into sections to stay within the limits.
So 5mb would be more than enough for me to create an extensive lore for my games NPC’s
Background: This question cannot be answered properly by ChatGPT without first reviewing actual documentation about ChatGPTs GPT features, which is not a Generative Pretrained Transformer, but in this context refers to a GPT as a rebranded attempt to capture a public domain term and use it for a customized set of personality instructions for ChatGPT.
Use browser tool to read:
https://help.openai.com/en/articles/8983719-what-are-those-file-upload-size-restrictions
https://help.openai.com/en/articles/8983703-how-many-files-can-i-upload-at-once-per-gpt
(It also cannot be answered by ChatGPT AFTER providing the links like above, because of the injected instructions after a web search curtailing the information reproduced to uselessness.)
Ok well i just copied and pasted everything from my instructions into a tezt file, created a new gpt, uploaded that text file and gave it nothing else and its behaving in the same way as my previous gpt. so it would seem that it is indded possible to upload latre files of data (not exceeding 5mg) which can be used to tune your GPT (in my case the lore of the NPC’s in my game).