Azure OpenAI 101: An introduction to Building Custom AI Models
Ғылым және технология
--------------- General Concepts ---------------
0:00 | Introduction
0:40 | GPT-3/Codex/DALL·E
1:30 | Roadmap of video
2:35 | Creating the Azure OpenAI resource in your Azure portal
3:55 | Azure OpenAI Studio
3:56 | Features overview & application examples: summarization, classifying text, natural language to SQL, Generating new product names
7:58 | GPT-3 Playground overview
8:45 | Model deployments & how to create
9:05 | Models & considerations
9:23 | Model naming convention
9:59 | Overview of GPT-3 models and capabilities: Davinci, Babbage, Ada, Curie
11:35| Overview of Codex models and capabilities: Cushman & Davinci
12:05| Recommendation for initial model deployment
------------------ Customizing Models ------------------
13:05| Generating python snippet from the GPT-3 playground
14:00| Defining the parameters to tune: Temperature, Max length tokens, top probabilities, frequency penalty, presence penalty, best of, pre-response text, post-response text
19:25| Scenarios on how to adjust model parameters
19:31| Scenario 1: low temperature, high top probability
21:06| Scenario 2: high temperature, high top probability, low frequency penalty
22:35| Scenario 3: moderate temperature, moderate top probability, low presence penalty
--------------- Fine-tuning using the OpenAI API in Python ---------------
24:15| Fine-tuning and use case, why would you want to fine-tune the model
24:42| Considerations: model size and its impact on computation and cost
25:20| Prompts and completions
25:50| Generating prompts and completions using the MediaWiki API to generate random titles and responses for those titles in Wikipedia
26:33| Importing necessary libraries
27:08| creating 100 random responses and writing a function to get their summaries, training data format
32:35| Training and validating datasets in JSON lines & exporting datasets to file management in Azure OpenAI
34:00| Creating customized model to fine-tine existing base models
35:30| Summary
--------- Documentation -----------
learn.microsoft.com/en-us/res...
learn.microsoft.com/en-us/azu...
learn.microsoft.com/en-us/azu...
Пікірлер: 27
that was a great video. good idea to update your video explaining the concepts with latest GPT-4 models
This was excellent bro, thanks for putting this together
Felt great watching this
Nice video, Abdul! Very thorough and detailed explanation on how to train custom OpenAI model. Thanks!!!
@abdulzedan
11 ай бұрын
Really happy to hear that it’s helped you! Thank you for the kind words ☺️
Awsome
Great
Open AI🔥🔥
Hi, Great video, I wanted ask that cant we directly give the contents of the file in fine tuning. If yes then what are the efforts for that and what will be its limitations. If the pdfs are to be needed to be used for long time. And what will happen if i want to add additional pdf contains after fine tuning a gpt model will i be able to do it ? Note I have also viewed your "Azure OpenAI 101: Powering ChatGPT with your Data - A Deep Dive" video but i want to know from this perspective. Thanks
Is there a timeline for when we should expect ChatGPT to be incorporated in Azure OpenAI?
@abdulzedan
Жыл бұрын
While the definitive timeline hasn't been announced just yet, it is noted to come soon. I recommend subscribing to the Microsoft Azure blog for announcements on this!
The video is very informative, but can we fine tune the codex models where we can train a new language to the Chatbot ?
@abdulzedan
10 ай бұрын
Thank you for the kind words - you can definitely do this, but it’s prudent to first see the available capabilities (I.e using the existing models with your data, I’ll be coming up with a video soon on that)! Cheers
How do you customise a model using your own data? In my case I have a 500pages of pdf file. Thanks!
@abdulzedan
8 ай бұрын
kzread.info/dash/bejne/jGqazZOgis2qfso.html
When you fine tune the model, do you get to play around with the epochs, learning rate etc ?
@abdulzedan
Жыл бұрын
A list of all the hyperparameters you can modify when training your model can be found here: learn.microsoft.com/en-us/rest/api/cognitiveservices/azureopenaistable/fine-tunes/create?tabs=HTTP This does include setting things like n_epochs and the learning_rate_multiplier!
@karimz361
Жыл бұрын
@@abdulzedan Thanks!
Hello, thank you for this amazing video. However, when I tried the same script you presented in the Microsoft documentation, I wasn't able to fine-tune my custom model and ended up with the message below: "" Job not in terminal status: notRunning. Waiting. Status: notRunning Status: failed Checking other fine-tune jobs in the subscription. Found 2 fine-tune jobs. "" Please, so you have any idea about how to fix this issue?
@abdulzedan
11 ай бұрын
I am glad you found this useful! Currently, fine-tuning is disabled for new customers in all regions, except for those who already had fine-tuning deployments. I would reach out to Azure Support to confirm that this is the issue for your job failure!
FYI, Davinci can no longer be fine tuned on Azure for new customers
@abdulzedan
Жыл бұрын
Thanks for the comment David. Currently, we aren't supporting any additional fine-tuning requests. However, if this wasn't the case, customers can open a support ticket to request fine-tuning for models (including Davinci)!
Great video! I’m looking to integrate Open AI service in my org and would love to speak about a potential opportunity for contract work. I didn’t see an email on your channel but if you’re interested lmk!
@abdulzedan
11 ай бұрын
I am glad you found this video helpful! Please reach out to me on LinkedIn if you have any inquiries!
# til
How much money are we talking about, in fine tuning things?
@abdulzedan
8 ай бұрын
Can't really give you a range as it can depend on a lot of factors. Here is a good article to check out: medium.com/devrain/calculating-azure-openai-service-usage-costs-a-comprehensive-guide-40b0880660f9