> For the complete documentation index, see [llms.txt](https://pixelfreund.gitbook.io/prompt-engineering/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://pixelfreund.gitbook.io/prompt-engineering/custom-gpts/faq.md). # FAQ ### How does model learning in custom gpts in chatgpt work exactly? Custom GPTs in ChatGPT do not actually perform any model learning or training. They are pre-configured AI assistants that leverage OpenAI's existing language models, such as GPT-3 or GPT-4, to provide specialized capabilities\[1]\[3]. The process of creating a Custom GPT involves: 1. Defining the purpose, instructions, and capabilities of the AI assistant using natural language prompts\[2]\[3] 2. Optionally providing additional knowledge sources, such as documents or datasets, to enhance the GPT's expertise in a particular domain\[2] 3. Configuring the GPT's abilities, such as web browsing, image generation, or code execution\[3] However, the underlying language model itself is not retrained or fine-tuned when creating a Custom GPT. The model's parameters remain fixed, and the GPT's specialized behavior is achieved through carefully crafted instructions and prompts\[2]\[3]. In summary, Custom GPTs in ChatGPT provide a user-friendly interface to leverage pre-trained language models for specific use cases, without requiring any actual model training or learning on the user's part. Quellen \[1] ChatGPT Custom GPT Models: Tailoring AI for Specific Needs \[2] How to create custom GPTs in ChatGPT - Pluralsight \[3] How to Make Custom ChatGPT Models: 5 Easy Steps to ... - DataCamp \[4] Introducing GPTs - OpenAI ### What File types are foreseen in openai custom gpt? According to the search results, OpenAI Custom GPTs support a variety of file types for knowledge sources and file uploads: 1. Knowledge sources can include many file types, such as PDF, text, CSV, and other data formats\[1]\[2]. These files are used to extend the GPT's knowledge with domain-specific information. 2. For file uploads, all common file extensions for text files, spreadsheets, presentations, and documents are supported\[3]. The maximum file size is 512MB per file, with a 2M token limit for text and document files (this limitation does not apply to spreadsheets)\[3]. 3. Images can also be uploaded, with a size limit of 20MB per image\[3]. However, images embedded in documents or presentations are not yet supported\[3]. 4. Files uploaded as knowledge to a Custom GPT are retained indefinitely, while files used for other purposes (e.g., Advanced Data Analysis) are retained for a duration that varies based on the user's plan\[3]. In summary, OpenAI Custom GPTs support a wide range of file types, including text files, spreadsheets, presentations, documents, and images, which can be used to extend the GPT's knowledge and capabilities for specific use cases. Quellen \[1] OpenAI Custom GPTs: What You Need to Worry About \[2] OpenAI GPT-3 API: Which file formats can be used for fine-tuning? \[3] File uploads FAQ - OpenAI Help Center \[4] Introducing GPTs - OpenAI ### What are maximum sizes in number of files and file size limits Based on the search results, the key file size and upload limits for OpenAI Custom GPTs are: 1. Maximum number of files per GPT: 20 files\[3] 2. Maximum file size per file: 512 MB\[2] 3. Maximum total file size across all files uploaded to a GPT: 100 GB per organization\[2] 4. Maximum token count for text/document files (excluding spreadsheets): 2 million tokens per file\[2] 5. Maximum image file size: 20 MB per image\[2] The search results also indicate that images embedded within documents or presentations are not yet supported\[2]. Additionally, the files uploaded as knowledge sources to a Custom GPT are retained indefinitely, while files used for other purposes (e.g., Advanced Data Analysis) are retained for a duration that varies based on the user's plan\[2]. So in summary, the key limits are 20 files per GPT, 512 MB per file, and a 100 GB total per organization, with some additional restrictions on token counts and image sizes. Quellen \[1] Number of files you can upload on a GPT - OpenAI Developer Forum \[2] File uploads FAQ - OpenAI Help Center \[3] GPTs knowledge capacity limits - OpenAI Developer Forum \[4] What is the max amount of data an Azure OpenAI Gpt-4 model can ... ### What is a token and how can it be calculated? A token represents the smallest meaningful unit of text in the context of large language models like GPT-3 and GPT-4\[2]\[4]. It can be a single character, a group of characters, a word, or even a punctuation mark\[3]. To calculate the number of tokens in a given text, you can use a tokenizer tool or follow these general guidelines: * In English, 1 token is approximately equal to 4 characters or 0.75 words\[2]\[4] * 100 tokens are roughly equivalent to 75 words\[2] * 1-2 sentences contain around 30 tokens, while 1 paragraph is about 100 tokens\[2] * 1,500 words are approximately equal to 2,048 tokens\[2] The exact number of tokens can vary depending on the language and the tokenization method used\[2]\[4]. For example, in Spanish and French, 1 word is about 2 tokens\[3]. Some key points about tokens: * Punctuation marks and special characters count as individual tokens\[3] * Emojis range from 2 to 3 tokens\[3] * The tokenization process can split words into sub-word tokens based on context\[2] * Newer models like GPT-4 use different tokenizers than previous versions, resulting in varying token counts for the same input text\[2] To get an accurate token count for your text, you can use online tools like the GPT-4 Token Counter\[3] or the Tokenizer tool provided by OpenAI\[2]. Quellen \[1] Token Counter \[2] What are tokens and how to count them? - OpenAI Help Center \[3] GPT-4 Token Counter Online - Quizgecko \[4] What Are Tokens and How to Calculate Them - LinkedIn ### What is a token and how can it be calculated? A token represents the smallest meaningful unit of text in the context of large language models like GPT-3 and GPT-4\[2]\[4]. It can be a single character, a group of characters, a word, or even a punctuation mark\[3]. To calculate the number of tokens in a given text, you can use a tokenizer tool or follow these general guidelines: * In English, 1 token is approximately equal to 4 characters or 0.75 words\[2]\[4] * 100 tokens are roughly equivalent to 75 words\[2] * 1-2 sentences contain around 30 tokens, while 1 paragraph is about 100 tokens\[2] * 1,500 words are approximately equal to 2,048 tokens\[2] The exact number of tokens can vary depending on the language and the tokenization method used\[2]\[4]. For example, in Spanish and French, 1 word is about 2 tokens\[3]. Some key points about tokens: * Punctuation marks and special characters count as individual tokens\[3] * Emojis range from 2 to 3 tokens\[3] * The tokenization process can split words into sub-word tokens based on context\[2] * Newer models like GPT-4 use different tokenizers than previous versions, resulting in varying token counts for the same input text\[2] To get an accurate token count for your text, you can use online tools like the GPT-4 Token Counter\[3] or the Tokenizer tool provided by OpenAI\[2]. Quellen \[1] Token Counter ; \[2] What are tokens and how to count them? - OpenAI Help Center ; \[3] GPT-4 Token Counter Online - Quizgecko ; \[4] What Are Tokens and How to Calculate Them - LinkedIn I ###