Thanks! A very nice demonstration. Wondering how you would access Amazon Inferentia from a Lambda function? My understanding is that Lambda cannot directly run inference on a GPU.
Wonderful walkthrough. So setting up an EFS within an EC2, calling lamba with pre-trained ML model, and then defining a trigger. My question is suppose you are continuously updating the model, what would you recommend to have updated trained models to refer to in the lambda? Let's say you are updating the model weekly. Do you think a viable solution is to set up a separate EC2 chron job, that triggers a similar layout (EC2/ EFS/ Lambda) but a TRAIN model lambda, that then outputs model to your other lambda's model dir? Then when you load model you use a maxdate model to always default to most recent?
Thanks for the video, it was an informative and explanative video on the subject. Still, I hava question tho: I believe Lambda uses only CPU runtime to make inference. Is there a way to use the same architecture and use a GPU runtime instead of a CPU one for more complicated models requiring GPU?
Try to do this, but its too much difficult. I stuck where aws lambda pick some of my package from EFS and not pick rest. For example abc is main package abd abc use xyz, so lambda pick abc and say xyz not found. Though both sre there
Short, precise and beautifully explained
Thanks! A very nice demonstration. Wondering how you would access Amazon Inferentia from a Lambda function? My understanding is that Lambda cannot directly run inference on a GPU.
Yes. Lambda functions don't have access to AI accelerators.
Wonderful walkthrough. So setting up an EFS within an EC2, calling lamba with pre-trained ML model, and then defining a trigger. My question is suppose you are continuously updating the model, what would you recommend to have updated trained models to refer to in the lambda? Let's say you are updating the model weekly. Do you think a viable solution is to set up a separate EC2 chron job, that triggers a similar layout (EC2/ EFS/ Lambda) but a TRAIN model lambda, that then outputs model to your other lambda's model dir? Then when you load model you use a maxdate model to always default to most recent?
hy did u try this out - regarding iam role code u have any links to share please help
awesomely explained..life saver
Thanks for the video, it was an informative and explanative video on the subject. Still, I hava question tho: I believe Lambda uses only CPU runtime to make inference. Is there a way to use the same architecture and use a GPU runtime instead of a CPU one for more complicated models requiring GPU?
Try to do this, but its too much difficult. I stuck where aws lambda pick some of my package from EFS and not pick rest. For example abc is main package abd abc use xyz, so lambda pick abc and say xyz not found. Though both sre there
can you please share the code base ? Thanks
Beautiful! Thanks
nice presenation..thanks
wow you rock sir, please make more videos
Can you please share the code