This content is fire! I love the efficiency of information delivery, slow enough to understand every step and speedy enough not to waste time, with visual helpers to focus on the right visual information. I also love that you kept errors in the video, so that we learn the most common we might encounter and how to tackle them. Keep it up!
Exceptional video! Thank you Danil! I watched tons of other videos and have no idea on how to deploy my ML model. Stumble on this one and now I'm able to do it.
Thanks a lot! Now I understand the core concepts. If I understand correctly, in Azure AI/ML, once a model is deployed, it provides a URL for inference along with a Docker image. I guess this is what the cloud provider is doing in the background - wrapping the model (e.g., the .joblib file) into a REST API (using something like Flask) and deploying it as a container in a Kubernetes (K8s) cluster. I guess something is happening with AWS Sage Maker
I don't know what exactly does Azure do when you deploy a model with them. In our example we didn't discuss inference scaling/load balancing. Azure probably has that handled out of the box. But the core concept of model deployment should be the same.
Thanks for the video and all the effort it took to produce it. It's easy to follow and nicely animated. 👍🏼 However, as of today I would rather use BentoML instead of FastAPI. It's even easier to set up an API and then publish a docker image.
Very nice video and clearly explain Currently i am learning about ci/cd and cloud deploy for ml project, could you kindly please do a video on that subject?
Thanks, I would to ask you, If i have a fine-tune trained model (with 20 GB size) then i follow the same steps in the video to make the api, and the docker steps, My question is what the best hosting provider to run the previous codes on the cloud? I need the previous pre training model not public (private), because it fine-tuning on sensitive company data. I hope you understand my question. Thanks 🙏❤
Yes, pickle is just a different container. The only thing to change in this case is the model loading part with pickle instead of joblib. The rest will be the same.
AWS or GCP, Azure i guess, every company now require every machine learning engineer to know about CI/CD pipeline, have experience with cloud service, or at least that is my point of view from recently job interview that i got deny :>> Gotta learn a lot lot more
I haven’t touched a windows machine is a decade, so can’t be certain, but I’m pretty sure that some modifications will be necessary. You may have to change the slashes direction in all the paths.
I know that the problem exists, but I’ve never encountered it. In my case Docker had been working perfectly fine. I guess compatibility issues arise only with certain OS versions.
@@thevicky1428 Just like any other model. Write the inference script to query the model with prompts or whatever you want to query it with, configure docker as explained in the video, save all the required llama artifacts into the corresponding directories and there you go. Basically repeat all the steps from the video only replacing the 'predict()' function with your llama inference code.
Nicely explained. I just wanted this much info only, saved myself from taking a 3-hr coursera course. Thank you
Glad it was helpful!
Yeah same here :))
This content is fire!
I love the efficiency of information delivery, slow enough to understand every step and speedy enough not to waste time, with visual helpers to focus on the right visual information.
I also love that you kept errors in the video, so that we learn the most common we might encounter and how to tackle them. Keep it up!
Thanks for the motivation!
Exceptional video! Thank you Danil!
I watched tons of other videos and have no idea on how to deploy my ML model. Stumble on this one and now I'm able to do it.
You're welcome! More cool videos coming soon
The most underrated channel i have ever seen , you give a very nice content and information with a very simple way , thank you very much
Thank you!
Very nicely explained!! Straight to the point..
Thanks a lot Danil.. You saved 4hrs of time. Its working for me :)
Glad it helped!
Thanks, Danil ! This is exactly what I was looking for. Clear and concise tutorial,🙏
Glad it was helpful!
Subscribed immediately. What a straight to point video. Thanks, man.
Thanks! More cool videos coming soon
Hello. This helped a lot. Good job Danil
Glad it helped
Thanks a ton!
No BS or extra. To the point as it needs to be.
Thanks a lot!
Now I understand the core concepts. If I understand correctly, in Azure AI/ML, once a model is deployed, it provides a URL for inference along with a Docker image. I guess this is what the cloud provider is doing in the background - wrapping the model (e.g., the .joblib file) into a REST API (using something like Flask) and deploying it as a container in a Kubernetes (K8s) cluster.
I guess something is happening with AWS Sage Maker
I don't know what exactly does Azure do when you deploy a model with them. In our example we didn't discuss inference scaling/load balancing. Azure probably has that handled out of the box. But the core concept of model deployment should be the same.
Thank you so much! This short but concise!
Thanks for the video and all the effort it took to produce it. It's easy to follow and nicely animated. 👍🏼
However, as of today I would rather use BentoML instead of FastAPI. It's even easier to set up an API and then publish a docker image.
Well, it's a matter of personal preference. I just got around Streamlit first...
precise tutorial ever😍
quick and clear... good job buddy
Glad it helped
This is just wonderful and succinct. Thank you!
Thank you for watching!
Very nice video and clearly explain
Currently i am learning about ci/cd and cloud deploy for ml project, could you kindly please do a video on that subject?
Coming up shortly.
This is awesome. Thank you for posting
Thank you for watching!
Thanks,
I would to ask you,
If i have a fine-tune trained model (with 20 GB size)
then i follow the same steps in the video to make the api, and the docker steps,
My question is what the best hosting provider to run the previous codes on the cloud?
I need the previous pre training model not public (private), because it fine-tuning on sensitive company data.
I hope you understand my question.
Thanks 🙏❤
@Danil_Zherebtsov
please create video on roadmap of mlops, and also end-to-end mlops projects, with and without open-source tools projects
Thanks for the comment. I’ll consider this
@@lifecrunch please upload fast as soon as possible, eagerly waiting here
thanks bro you explain so good God with you !
Happy to help
i bild a object detection model . that was 180mb in size . how can i deploy my model
That’s an open ended question. Deploy where? I have a few videos on the subject, check them out.
Nicely explained, please will it be the same steps for a pikl file ?
Yes, pickle is just a different container. The only thing to change in this case is the model loading part with pickle instead of joblib. The rest will be the same.
How or where can we deploy that Docker container to be used along the internet?
AWS or GCP, Azure i guess, every company now require every machine learning engineer to know about CI/CD pipeline, have experience with cloud service, or at least that is my point of view from recently job interview that i got deny :>> Gotta learn a lot lot more
@@piano_tam97106 I'm sorry to hear that, thanks for the reply, good luck!
I guess I need to make a separate video on this subject. Stay tuned.
Very informative.
Thanks!
is it possible i do a deployment with model.RData in R instead of model.joblib
If your model inference will run R as the inference environment then yes
Thanks a lot for the great video. Somehow the links for course 3 and 4 are invalid. Could you please help update the links?
Updated. The problem was that courses 3 and 4 were merged together and had a new link. I've posted the updated link.
I am getting an Aborted ! Error whenever i am using the docker --run command. Anyone knows whats going on ?
Did you install docker correctly?
Can we deploy for the code that is written in jypter notebook
Jupyter notebook is a research environment, not development. For production applications use .py files.
The commands in terminal work in os Windows?
I haven’t touched a windows machine is a decade, so can’t be certain, but I’m pretty sure that some modifications will be necessary.
You may have to change the slashes direction in all the paths.
Do you know if the problem of compatibility between M chips and Docker is solved?
I know that the problem exists, but I’ve never encountered it. In my case Docker had been working perfectly fine.
I guess compatibility issues arise only with certain OS versions.
can i deploy my computer vision project using this method
Sure. You can deploy any model this way.
cheers mate thanks!
Any time!
Can I do llama 8b fine tuned with this sir ?
With what? Docker??
@@lifecrunch yeah, I fine tuned unsloth llama 8.1 how to deploy that with docker or cloud providers
@@thevicky1428 Just like any other model. Write the inference script to query the model with prompts or whatever you want to query it with, configure docker as explained in the video, save all the required llama artifacts into the corresponding directories and there you go. Basically repeat all the steps from the video only replacing the 'predict()' function with your llama inference code.
@@lifecrunch thanks sir
Thank you !
You're welcome!
Can we do the same with open source model?
What do you mean by open-source model?
If you have any trained model and code to inference it - you can deploy it.
Thank you for your reply, i think I got it.
very helpful !!!!!!!!!!!
Glad it helped!
Thanks Sir
Welcome!
👑🙌🙌
❤❤❤❤❤❤❤
Seems Hard)
The intention was to show how easy it actually is 😉
soo good
Thanks man!
remove background audio track its distracting
Thanks for the feedback. Unfortunately, published videos cannot be amended.
Дружище, давай на русском. Больше ведь будет просмотров
Это как ты так посчитал?) Русскоязычное население 220 млн., а на английском говорят 1.4 млрд…
@@lifecrunch на международном английском говорят 1.4 млрд, но просмотры твои на международный уровень как-то не тянут. Не все математикой объясняется)
/bin/sh: 1 : [uvicorn,: not found
A little more context wouldn’t hurt
"PromoSM" 🤤
There is nothing to promote here. Common practice.
Thanks!
Welcome!