For the ones who struggle to install with timeouts, this lifecycle script will only work if it doesnt take longer than 5min. Julien used a quiet heavy instance type, this lifecycle script wont work on smaller instances, and gets an timeout after 5min. But you can install it manualy to fix this issue
@@juliensimonfr I have to say I love your videos, specialy that you render in 60fps, much more smooth to watch, specialy in coding should be a must. Could you create maybe a full tutorial on how to finetune lets say LLama2 on text like xml oder JSON or so and maybe show the full process of getting a model production ready behind the AWS inference endpoint with multiple instances, to get low responsetimes :D
This is awesome! I wonder if it's possible to use local vscode and its ssh remote host function to connect to the sagemaker notebook so that I can keep all the settings and extensions on my vscode.
Thanks for sharing, Julien. I'm getting up to speed on using SageMaker Studio, and it's an adjustment being used to just using VSCode locally for software or data engineering and then deploying code to the cloud through its console. It seems to me that some people use notebooks in SageMaker almost as a CLI to not only experiment and explore the data, but to also output the scripts for executing in production. What are some alternate workflows or best practices for ML engineering that you have seen in your experience or amongst your colleagues at Huggingface? By the way, I am very inclined to try out this approach. Notebooks as an IDE in SageMaker just doesn't feel right to me.
Thanks for making this video. What do you use for this set up instead of Pylance? I'm really missing the ability to ctrl-click or mouseover a variable and go to the object definition.
Thank you for the workshop-it was exactly what I needed. I do have one question though: if I want to save money and use my local machine to train the model, is there an easy way to quickly switch from AWS ec2 instance to my local machine?
Thank you. Yes, you can run the SageMaker SDK locally on your machine. The main difference is how to setup credentials with IAM, see See ua-cam.com/video/K3ngZKF31mc/v-deo.html
Login with SageMaker Studio, installed the extension ok and login as well ok. But then. [INFO] [auth] [2023-02-15T22:14:49.339Z] Invalid copilot token: missing token: 403 [ERROR] [default] [2023-02-15T22:14:49.342Z] GitHub Copilot could not connect to server. Extension activation failed: "No access to GitHub Copilot found." Any tip?
Thanks for the instructions. I followed them successfully, but the CoPilot plugin doesn't work with Code Server. Is there some kind of workaround to get it to work?
Thanks for this video! I managed to reproduce to steps, but for some reason I can not edit anything inside the launched VScode, I get the error: NoPermissions (FileSystemError): Error: EACCES: permission denied.. Any ideas why this might be happening?
I guess not. Maybe you could set up with a fancy networking setup to access instances in a private VPC (through a DMZ or something), but I've never seen any example of that.
i don't know why UA-cam didn't recommend this channel long before. youtube needs to recommend based on quality rather than merely popularity.
Couldn't agree more :)
For the ones who struggle to install with timeouts, this lifecycle script will only work if it doesnt take longer than 5min.
Julien used a quiet heavy instance type, this lifecycle script wont work on smaller instances, and gets an timeout after 5min.
But you can install it manualy to fix this issue
Yes, good point!
@@juliensimonfr I have to say I love your videos, specialy that you render in 60fps, much more smooth to watch, specialy in coding should be a must.
Could you create maybe a full tutorial on how to finetune lets say LLama2 on text like xml oder JSON or so and maybe show the full process of getting a model production ready behind the AWS inference endpoint with multiple instances, to get low responsetimes :D
This is awesome! I wonder if it's possible to use local vscode and its ssh remote host function to connect to the sagemaker notebook so that I can keep all the settings and extensions on my vscode.
I do that all the time with plain EC2 instances, it works great.
Thanks for sharing, Julien. I'm getting up to speed on using SageMaker Studio, and it's an adjustment being used to just using VSCode locally for software or data engineering and then deploying code to the cloud through its console. It seems to me that some people use notebooks in SageMaker almost as a CLI to not only experiment and explore the data, but to also output the scripts for executing in production. What are some alternate workflows or best practices for ML engineering that you have seen in your experience or amongst your colleagues at Huggingface? By the way, I am very inclined to try out this approach. Notebooks as an IDE in SageMaker just doesn't feel right to me.
Thanks for making this video. What do you use for this set up instead of Pylance? I'm really missing the ability to ctrl-click or mouseover a variable and go to the object definition.
Thanks for sharing Julien.
My pleasure!
this is awesome but the copilot extension no longer works. if you are still using this workflow, do you have a workaround?
Thank you for the workshop-it was exactly what I needed. I do have one question though: if I want to save money and use my local machine to train the model, is there an easy way to quickly switch from AWS ec2 instance to my local machine?
Thank you. Yes, you can run the SageMaker SDK locally on your machine. The main difference is how to setup credentials with IAM, see See ua-cam.com/video/K3ngZKF31mc/v-deo.html
Login with SageMaker Studio, installed the extension ok and login as well ok. But then.
[INFO] [auth] [2023-02-15T22:14:49.339Z] Invalid copilot token: missing token: 403
[ERROR] [default] [2023-02-15T22:14:49.342Z] GitHub Copilot could not connect to server. Extension activation failed: "No access to GitHub Copilot found."
Any tip?
Thanks for the instructions. I followed them successfully, but the CoPilot plugin doesn't work with Code Server. Is there some kind of workaround to get it to work?
Thanks for this video! I managed to reproduce to steps, but for some reason I can not edit anything inside the launched VScode, I get the error: NoPermissions (FileSystemError): Error: EACCES: permission denied.. Any ideas why this might be happening?
SageMaker only lets you write in /home/ec2-user/SageMaker. You're probably trying to write elsewhere.
this is awesome. just wondering can we also run code-server on google colab environtment?
Thanks. I don't use Colab, but if someone knows, please leave a comment :)
Terminal commands that does training (like RASA train) are killed automatically in sagemaker code server
Can we use this without enabling internet access to Sagemaker Instance ?
I guess not. Maybe you could set up with a fancy networking setup to access instances in a private VPC (through a DMZ or something), but I've never seen any example of that.