Autogen Tut

RIP to the milkman, elevator operator, and the guy who walked around waking people up before alarm clocks.

Final Project Work

A2 DLNN

CS559 Notes

Explore

Final Project Work

Eddie Coda

Last edited 158 days ago by System Writer

Literature Review

Overall Project Architecture & Procedural Steps

1.
`Tensordock`⁠
VM Setup and Configuration

Running the instance - Login here:

⁠

https://console.tensordock.com/home⁠

⁠

Username:

⁠

eddiecoda24@gmail.com⁠

⁠

Password:

Xthequickbrownfox321.

Find "List Servers" && click it.

Click on the yocoda server

Go to the actions tab && start

If you go back to the overview tab, there are some instructions

If you go to the network tab, you will get IP address.

⁠

2. Jupyter Lab Configuration Once VM is Running

In local terminal:

ssh user@216.153.52.16

Password:

Xthequickbrownfox321.

Run:

jupyter lab --ip=0.0.0.0 --no-browser

for a Jupyter Lab or:

jupyter notebook --ip=0.0.0.0 --no-browser

for a Jupyter Notebook

After starting Jupyter Lab, it will provide a URL in the terminal, typically something like http://127.0.0.1:8888/?token=xxxxxxxx. You need to replace 127.0.0.1 with your VM's IP address (216.153.52.16 in your case)

Back on your local machine:

⁠

http://216.153.52.16:8888/lab?token=9e8eabfaf09ca9dc1d3711e20c181542d35bf36c0c8dbadd⁠

⁠

Launch the Jupter lab, if prompted for the token you can find that on the API page of the Your Account tab.

⁠

http://216.153.52.16:8888/lab⁠

⁠

4. Version Control with Git

Step 1: Set Up SSH Keys for GitHub

First, you need to set up SSH keys for GitHub on your VM. This will allow secure communication between your VM and GitHub.

Generate a new SSH key pair on your VM using the ssh-keygen command. You can use the following command to generate a new key pair:

When prompted, press enter to accept the default file location, and enter a secure passphrase

2⁠

Once the key pair is generated, you need to add the public key to your GitHub account. First, display your public key with the following command:

Copy the output, then go to GitHub, navigate to your account settings, click on the SSH and GPG link, and click Add Key to register the public SSH key with your account

2⁠

Step 2: Clone Your GitHub Repository

Next, clone your GitHub repository to your VM. You can do this with the following command:

This will create a new directory named facebert containing your GitHub repository.

Step 3: Copy Files to the Repository Directory

Now, you need to copy the files you want to push to GitHub into the facebert directory. You can do this with the cp command. For example, to copy all files from the workspace directory, you can use the following command:

Step 4: Push Files to GitHub

Finally, you can push the files to GitHub. Navigate to the facebert directory and use the git add, git commit, and git push commands to add the files to the repository and push them to GitHub:

This will add all files in the facebert directory to the repository, commit the changes with the message "Add files from VM", and push the changes to the master branch of your GitHub repository

1⁠

⁠

3⁠

⁠

7⁠

Please replace "master" with the appropriate branch name if your default branch is not named "master".

Navigate to your project directory:

Pull the Latest Changes:

Replace main with the branch name you want to update, if it's different.

Resolve Any Merge Conflicts:

If there are conflicts, you'll need to manually resolve them.

Check the Status:

After the pull, confirm everything is up to date and integrated.

Optional - Push Local Changes:

If you have made local changes to push.

Using git pull is more straightforward if you are comfortable with automatically merging changes without reviewing them first. Remember, git fetch and git merge give you more control over the process by allowing you to review changes before integrating them.

5. Data Acquisition and Preprocessing

Dataset Management

Download the required datasets directly to the VM.

Perform any necessary preprocessing steps, leveraging the VM's computational resources.

6. Integration with Weights & Biases
`(W&B)`⁠
⁠

W&B Configuration

Set up W&B on the Tensordock VM to track your experiments.

Ensure W&B is configured to log training metrics, hyperparameters, and evaluation results.

7. Model Training and Fine-Tuning

Model Access and Fine-Tuning

Access BERT, CrammedBert, and UltraFastBert via Hugging Face Transformers.

Fine-tune these models using your datasets, utilizing the VM's powerful GPUs for efficient training.

8. Model Deployment and Inference

Model Deployment

Deploy fine-tuned models either on Tensordock (if it supports hosting) or use Hugging Face for model deployment and access via inference endpoints.

9. Evaluation with FACE Metric

Applying FACE Evaluation

Evaluate the performance of your models using the FACE metric.

Use Tensordock's computational power to quickly process these evaluations.

10. Performance Analysis and Reporting

Analysis Using W&B

Utilize W&B for in-depth analysis and visualization of model performance.

Generate reports and insights based on the collected data.

11. Resource Management and Monitoring

Efficient Use of VM Resources

Monitor the usage of your VM resources to ensure efficient operation.

Adjust your usage based on the project's computational demands and your time constraints.

Overall Coding Flow

Initial Setup:

Creating the project structure.

create folders for dataset

create folders for scripts & notebooks

Setting up the initial environment and dependencies.

Hugging face setup with either:

pytorch

pytorch lightning

transformers

Data Acquisition and Preprocessing:

Writing scripts for downloading and preprocessing the data.

Model Setup and Training:

Coding the model initialization, training loops, and logging.

Model Evaluation:

Implementing the evaluation phase using the FACE metric.

Result Analysis and Visualization:

⁠

https://docs.wandb.ai/guides/sweeps/walkthrough⁠

⁠

Scripts for analyzing results and creating visualizations.

Regular Updates and Documentation:

Committing changes to Git and documenting the process.

rando

Want to print your doc?
This is not the way.

Try clicking the ⋯ next to your doc name or using a keyboard shortcut (

CtrlP

) instead.

Final Project Work

Overall Project Architecture & Procedural Steps

1. Tensordock⁠ VM Setup and Configuration

Running the instance - Login here:

Username:

Password:

2. Jupyter Lab Configuration Once VM is Running

In local terminal:

Password:

Run:

for a Jupyter Lab or:

for a Jupyter Notebook

Back on your local machine:

Launch the Jupter lab, if prompted for the token you can find that on the API page of the Your Account tab.

4. Version Control with Git

Step 2: Clone Your GitHub Repository

Step 3: Copy Files to the Repository Directory

Step 4: Push Files to GitHub

5. Data Acquisition and Preprocessing

6. Integration with Weights & Biases (W&B)⁠⁠

7. Model Training and Fine-Tuning

8. Model Deployment and Inference

9. Evaluation with FACE Metric

10. Performance Analysis and Reporting

11. Resource Management and Monitoring

Overall Coding Flow

rando

1.
`Tensordock`⁠
VM Setup and Configuration

6. Integration with Weights & Biases
`(W&B)`⁠
⁠