Windows AI Studio Preview
Overview
Windows AI Studio simplifies generative AI app development by bringing together cutting-edge AI development tools and models from Azure AI Studio Catalog and other catalogs like Hugging Face. You will be able browse the AI models catalog powered by Azure ML and Hugging Face, download them locally, fine-tune, test and use them in your Windows application. As all of the computation happens locally, please make sure your device can handle the load. In the future, we plan to integrate ORT/DML into the Windows AI Studio workflow so developers can run AI models on any Windows Hardware.
Quick Start
In this section you will learn how to quickly start with Windows AI Studio.
Prerequisites
Windows AI Studio will run only on NVIDIA GPUs for the preview, so please make sure to check your device spec prior to installing it. WSL Ubuntu distro 18.4 or greater should be installed and is set to default prior to using Windows AI Studio. Learn more how to install Windows subsystem for Linux and changing default distribution.
Install Windows AI Studio Preview
Windows AI Studio is shipped as a Visual Studio Code Extension, so you need to install VS Code first, and download Windows AI Studio from the VS Marketplace.
Prerequisites Check
After the Windows AI Studio is installed, we will check your system to make sure you have the right resources to run it. After prerequisites check is completed, you can complete the environment set up by clicking "Set Up WSL Environment" button.
Now, you are ready to use the extension! You will be prompted to sign in to GitHub, so please click "Allow" to continue. You will be redirected to GitHub signing page. Please sign in and follow the process steps. After successful completion, you will be redirected to VS Code.
Let's explore the available actions!
Available Actions
Upon launching Windows AI Studio, you can select from the following options:
- Model Fine-tuning
- RAG Project
- Phi-2 Model Playground
- Windows Optimized Models
Model Fine-tuning
To initiate the local fine-tuning session using QLoRA, select a model you want to fine-tune from our catalog powered by AzureML.
Note You do not need an Azure Account to download the models
Start by selecting a project name and location. Next, select a model from the model catalog. You will be prompted to download the project template. You can then click "Configure Project" to adjust various settings. We use Olive to run QLoRA fine-tuning on a PyTorch model from our catalog. All of the settings are preset with the default values to optimize to run the fine-tuning process locally with optimized use of memory, but it can be adjusted for your scenario.
Settings | Data Type | Default Value | Description |
---|---|---|---|
Compute Dtype | Str | bfloat16 | Data type for model weights and adapter weights. For 4bit quantized model, it is also the computation data type for the quantized modules. Should be one of bfloat16, float16 or float32 |
Quant type | nf4 | Quantization data type to use. Should be one of fp4 or nf4 | |
Double quant | Bool | yes | Whether to use nested quantization where the quantization constants from the first quantization are quantized again |
Lora r | Int | 64 | Lora attention dimension |
Lora alpha | Float | 16 | The alpha parameter for Lora scaling |
Lora dropout | Float | 0.1 | The dropout probability for Lora layers |
Eval dataset size | Float | 1024 | Size of the validation dataset |
Seed | Int | 0 | Random seed for initialization |
Data Seed | Int | 42 | Random seed to be used with data samplers |
Per device train batch size | Int | 1 | The batch size per GPU for training |
Per device eval batch size | Int | 1 | The batch size per GPU for evaluation |
Gradient accumulation steps | Int | 4 | Number of updates steps to accumulate the gradients for, before performing a backward/update pass |
Enable Gradient checkpoint | Bool | yes | Use gradient checkpointing. Recommended to save the memory |
Learning rate | Float | 0.0002 | The initial learning rate for AdamW |
Max steps | Int | -1 | If set to a positive number, the total number of training steps to perform. Overrides num_train_epochs. In case of using a finite iterable dataset the training may stop before reaching the set number of steps when all data is exhausted |
After all the parameters are set, click Generate Project. This will:
- Initiate the model download
- Install all prerequisites and dependencies
- Create VS Code workspace
When the model is downloaded, you can launch the project from Windows AI Studio. You can now fine-tune and test the model by following the instructions from Readme page in the project folder.
RAG Project
Coming soon!
Phi-2 Model Playground
Coming soon!
Windows Optimized Models
This is the collection of publicly available AI models already optimized for Windows. The models are stored in the different locations including Hugging Face, GitHub and others, but you can browse the models and find all of them in one place ready for downloading and using in your Windows application.
Q&A
Please refer to our Q&A page for most common issues and resolutions
Contributing
This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit https://cla.opensource.microsoft.com.
When you submit a pull request, a CLA bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., status check, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA. This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.
Trademarks
This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft's Trademark & Brand Guidelines. Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party's policies.
from Hacker News https://ift.tt/WafrMz3
No comments:
Post a Comment
Note: Only a member of this blog may post a comment.