I have been building NLP applications on top of GPT3 and other Open Source Library for the last couple years. I am looking for interesting projects to work on that can be build on top of GPT, no fees.
Leave a comment or drop me a message here if you are interested.
32 Likes
Hi Nelson,
thank you for this proposal.
I want to develop some tools to help me in my SEO work, based on OpenAI.
I would like to create a kind of toolbox. I know the prompts I want to use.
I need a form, on a simple wordpress page, where I can fill in some parameters. From there, the prompt and its parameters are sent to OpenAI and then we get the response that is displayed in the same. I tried to describe my need to ChatGPT but the proposed extension does not work properly.
Can you help me?
Thanks
Questions for you @DigitalSun
- Is the toolbox for internal or external use?
- There is no official API for ChatGPT, but the Completions API is available
- Can you clarify the issue you have with the proposed extension?
4 Likes
HI @nelson, I’m looking to build a chat bot with GPT on top of custom documents. I went through fine tuning but that requires 500 question answer pairs which is not going to scale. Have you seen any solutions which take a bunch of docs to train the model?
2 Likes
- Is the toolbox for internal or external use?
external use
- There is no official API for ChatGPT, but the Completions API is available
yes i know, that’s the API that I try to use
- Can you clarify the issue you have with the proposed extension?
yes, i don’t receive the answer from OpenAI
Here is the code of the extension PHP for WP
<?php /* * Plugin Name: Extension OpenAI * Plugin URI: https://www.exemple.com * Description: Extension qui permet d'utiliser OpenAI par le biais de son API * Version: 1.0 * Author: John Doe * Author URI: https://www.exemple.com * License: GPL2 */ /* * Shortcode */ function openAI_shortcode(){ // Affichage du formulaire $html = ''; // Champ de la clé API $html .= 'Entrez votre clé API OpenAI
'; $html .= ''; // Champ du prompt Ă executer $html .= 'Entrez le prompt Ă executer
'; $html .= ''; // Bouton GO $html .= ''; // Affichage du résultat if( isset($_POST['openAI_go']) ) { $openAI_api_key = esc_attr($_POST['openAI_api_key']); $openAI_prompt = esc_attr($_POST['openAI_prompt']); // Appel API OpenAI $result = openAI_call($openAI_api_key, $openAI_prompt); // Affichage du résultat $html .= 'Résultat du prompt :
'; $html .= '' . $result . '
'; } $html .= ''; return $html; } add_shortcode('openAI', 'openAI_shortcode'); /* * Appel API OpenAI */ function openAI_call($openAI_api_key, $openAI_prompt){ // Requête $curl = curl_init(); curl_setopt_array($curl, array( CURLOPT_URL => "https://api.openai.com/v1/completions", CURLOPT_RETURNTRANSFER => true, CURLOPT_ENCODING => "", CURLOPT_MAXREDIRS => 10, CURLOPT_TIMEOUT => 200, CURLOPT_FOLLOWLOCATION => false, CURLOPT_HTTP_VERSION => CURL_HTTP_VERSION_1_1, CURLOPT_CUSTOMREQUEST => "POST", CURLOPT_POSTFIELDS => '{"prompt": "' . $openAI_prompt . '","max_tokens": 50,"temperature": 0.9,"top_p": 1.0,"n": 1,"stream": false,"logprobs": null,"stop": [" "],"engine": "davinci","frequency_penalty": 0.0,"presence_penalty": 0.0}', CURLOPT_HTTPHEADER => array( "Content-Type: application/json", "Authorization: Bearer " . $openAI_api_key ), )); // Récupération du résultat $response = curl_exec($curl); curl_close($curl); $response = json_decode($response); $result = $response->choices[0]->text; return $result; }Hi @kartik
I made created an integration for the GPT Embedding API and Completion API using custom data uploaded on Google Spreadsheet.
Will this do the job?
Yes this is pretty helpful but its not fully accurate and there’s no way to ask follow up questions. Also I was looking for a way to scale to 100s of documents/text. wouldn’t training the model be better here?
@kartik I see, I can take a look further. Can you message me a tiny sample of some text to look for and a couple follow up questions. Something simple, non confidential that can fit into 1 prompt request will be great.
Sure. Here you go -
It’s applicable to full time employees and Interns. Full time employees will receive Rs 50,000 for their WFH setup. Employees will receive Rs 25,000 on their first day of work. The additional Rs 25, 000 will be applicable once the employee completes 6 months in the company. After completing 1 year, the employee will receive Rs 5,000 on their work anniversary for their WFH setup. For Interns whose internship tenure is more than 6 months is eligible for WFH setup worth Rs 25,000.
Hey Nelson. Firstly thank you for being kind enough to “give back” and take some time to share your knowledge. Ive been messing around with OpenAi python api for the past 12-months — however Im struggling with 1 or 2 concepts which Im struggling to wrap my head around. Is it possible I could be selfish enough to request either: A 5-minute call or screen share session with you….
yes correct. What if you had to feed in hundreds of docs?
@timco sure, can you send me some sample code and the issue you are running to first, so I can take a better look.
@kartik are you familiar with the embeddings API, that can help you do search and use the most relevant information in your prompt. Are you using just the UI or are you using the API from your application?
yes using the API. I’m not familiar - will check that. How did you achieve the one in the screenshot? The follow up question?
Thank you.
@kartik In the worksheet I provided, you can use the Embeddings API to encode your existing data source to embeddings and then use a custom cosine similarity function in the spreadsheet to find the relevance. Give it a try, it that still doesn’t work, let me know and we can think of something else…
4 Likes
Hey @nelson I’m really interested to know how this embeddings Google sheet thing works, as it seems like it might be a really good solution to token limit problem.
I’d like to provide some specific context information with the prompt, but the supporting information is too large in size for the token limit.
My thinking is this might be a way around that, I’m curious how the cosine aspect of this works?
Hi @multitechvisions
OpenAI offers the Embeddings API with 8191 tokens, so let’s say you have document of with roughly 32,764 tokens, which is too big to fit inside a GPT prompt.
You can create divide the document into 20 records around 1639 tokens each and use this API to encode into embeddings. When you want to include one of this record in to your prompt, you can do a similarity search on the 20 embeddings using Cosine Similarity and find the most relevant record. After that, it’s just prompt engineering.
Since Google Sheet support custom functions, I just build the OpenAI APIs and Cosine Similarity into custom functions so anyone can chain together a series of API Request and Response without the need of deep programing skills.
The google sheet is free to use, let me know if you have any issues running it, I’ll be happy to help.
3 Likes
Yeah I’d like if u can help me with this. What’s your contact information