#mlengineering — Public Fediverse posts

One thing I am currently learning at my new job is that simple heuristics can often improve the performance of an ML system by a lot.

#ml #ai #mlengineering

Structured data drives AI. But messy inputs? They stall everything.
We’ve listed six parsing issues you should be watching for.
👉 Read the blog to know more: https://shorturl.at/vuJjw

#AIanalytics #MLengineering #DataWrangling #ParsingProblems #TechStrategy #BigData

Early in your ML career, every decision feels irreversible. But the best engineers don’t aim for perfection—they build with reversibility in mind.

Understanding the difference between one-way and two-way doors will help you iterate faster and build better.

#MachineLearning #MLEngineering #TechCareers

A few tips for optimizing Pytorch model training time from a Yandex ML engineer.

https://alexdremov.me/simple-ways-to-speedup-your-pytorch-model-training/

#ml #mlengineering #modeltraining #pytorch #modeloptimization

A few tips for optimizing Pytorch model training time from a Yandex ML engineer.

https://alexdremov.me/simple-ways-to-speedup-your-pytorch-model-training/

#ml #mlengineering #modeltraining #pytorch #modeloptimization

A few tips for optimizing Pytorch model training time from a Yandex ML engineer.

https://alexdremov.me/simple-ways-to-speedup-your-pytorch-model-training/

#ml #mlengineering #modeltraining #pytorch #modeloptimization

A few tips for optimizing Pytorch model training time from a Yandex ML engineer.

https://alexdremov.me/simple-ways-to-speedup-your-pytorch-model-training/

#ml #mlengineering #modeltraining #pytorch #modeloptimization

A few tips for optimizing Pytorch model training time from a Yandex ML engineer.

https://alexdremov.me/simple-ways-to-speedup-your-pytorch-model-training/

#ml #mlengineering #modeltraining #pytorch #modeloptimization

Here's a more clearly visible demonstration of the problem I described previously: https://sigmoid.social/@chrisoffner3d/111591367887994819

On the left we see the progression of cross-attention maps extracted via the CPU, on the right we see the same cross-attention maps extracted via the GPU.

This is using the #Keras implementation of #StableDiffusion on an M3 Max.

#TensorFlow #StableDiffusion #Diffusion #Python #MLEngineering #MachineLearning #DeepLearning #GPU #M3Max

Here's a more clearly visible demonstration of the problem I described previously: https://sigmoid.social/@chrisoffner3d/111591367887994819

On the left we see the progression of cross-attention maps extracted via the CPU, on the right we see the same cross-attention maps extracted via the GPU.

This is using the #Keras implementation of #StableDiffusion on an M3 Max.

#TensorFlow #StableDiffusion #Diffusion #Python #MLEngineering #MachineLearning #DeepLearning #GPU #M3Max

Here's a more clearly visible demonstration of the problem I described previously: https://sigmoid.social/@chrisoffner3d/111591367887994819

On the left we see the progression of cross-attention maps extracted via the CPU, on the right we see the same cross-attention maps extracted via the GPU.

This is using the #Keras implementation of #StableDiffusion on an M3 Max.

#TensorFlow #StableDiffusion #Diffusion #Python #MLEngineering #MachineLearning #DeepLearning #GPU #M3Max

Here's a more clearly visible demonstration of the problem I described previously: https://sigmoid.social/@chrisoffner3d/111591367887994819

On the left we see the progression of cross-attention maps extracted via the CPU, on the right we see the same cross-attention maps extracted via the GPU.

This is using the #Keras implementation of #StableDiffusion on an M3 Max.

#TensorFlow #StableDiffusion #Diffusion #Python #MLEngineering #MachineLearning #DeepLearning #GPU #M3Max

Here's a more clearly visible demonstration of the problem I described previously: https://sigmoid.social/@chrisoffner3d/111591367887994819

On the left we see the progression of cross-attention maps extracted via the CPU, on the right we see the same cross-attention maps extracted via the GPU.

This is using the #Keras implementation of #StableDiffusion on an M3 Max.

#TensorFlow #StableDiffusion #Diffusion #Python #MLEngineering #MachineLearning #DeepLearning #GPU #M3Max

For example, check the second row, fifth column and how it changes between t = 600 and t = 700.

Is this some bug specific to Apple GPUs or does this also happen with CUDA?

For t = 0, the CPU and GPU images look identical. For higher t, the GPU run produces *very* different results even when re-running with the exact same model inputs, i.e. also for the same time step t.

Any idea why that is?

#MLEngineering #GPU #DeepLearning #Diffusion #CUDA #AppleSilicon #TensorFlow #Keras

I'm running into some unexpected and significant non-determinism when running a #Keras diffusion model on my Apple GPU.

On the left we see the progression of cross-attention maps for time steps from t = 0 to t = 900 when running the model via the CPU.

We see that each cross-attention map undergoes some "refinement" progression as we go from t = 0 to t= 900.

On the right we see the same but on the GPU.

It's a much more erratic and discontinuous progression.

#MLEngineering #DeepLearning #GPU

🔖 The Top 5 Papers About #mlops You Should Know (Part 1)

I've seen a ton of lists about the most important papers in #ml, #datascience, #deeplearning, #mlengineering.

But I've either seen not that many #mlops reading lists or when I do run across them they tend to be focused a bit too deeply on specific ML systems or domains or algorithms.

👉🏻 If you only read 5 papers to understand why ML is hard (and how big the problem space of MLOps is) it should be these papers.

[To Be Continued]

Does anyone here have experience with #Prefect? What's the best way to automate blocks? can you do it via #terraform? #ml #mlengineering

The tools we have today are better than the ones we had before and this is especially true in the #mlops world. We have more options than ever before (cc: MAD Turck Landscape) but confusion is just as high as it ever was.

#mlops #productionml #mlengineering #oss #devtools #python

Having #DataScientists Build Infrastructure & Developing Models At The Same Time Is A Terrible Anti-Pattern We’re Addicted To.

Esp at comps that aren’t early stage -- correlated w/ a lack of technical DS leadership, poor infra design, and lack of organizational alignment.

Really shows how the difference between success & failure isn’t technology choices but good project management & strategic leadership around platforms.

#mlops #mlengineering #mlplatforms #datascience

👉🏻 Online Inference =/= Streaming

We're all aware of this right? That they're not the same thing?

#mlops #mlengineering #datascience #dataengineering #productionml #mlsystems #systemdesign

🤔 Rather than trying to get rid of the #datascientist title, maybe we just treat it as an abstract class and continue on our merry ways?

#datascience #dataengineering #mlops #mlengineering #ai #career #data

🤔To bootcamp or not to bootcamp?

Like all annoying senior devs, my answer is going to be: "It depends".

I breakdown what consider when choosing the #bootcamp route for #datascience (but advice good for other bootcamps like #dataengineering #mlengineering, etc)

#🐘 https://t.co/sTiiwWOB7D

If the answer is similar to:
1️⃣ ASAP
2️⃣ Minimal
3️⃣ Divorced
4️⃣ We can't
5️⃣ Less than 5

Then your first step shouldn't be building an ML platform, it should be developing models or ML-drive product features using the simplest, tried & true patterns possible.

#mlops #mlplatform #datascience #mlengineering #platformengineering #dataengineering #ai #mlinproduction

there’s a lot of really cool stuff in #MLEngineering that amounts to “train another model”. like using #SHAP to automate feature selection (first you have to train a model though). or #ConceptSHAP where you train simple linear models on the output of each neural net layer. or anomaly detection, or autoencoders, or...

😳 My talk proposal to the #mlops track was accepted to #DataCouncilAustin 2023 🤯

🎉 What an exciting way to start the year! 😃

Looking forward to connecting with folks in Austin from March 28-30th on #mlops #productionml #mlengineering #productiondatascience #DataCouncilAustin2023 #datacouncil

Please feel free to connect with me on LI if you're attending or presenting!
https://www.linkedin.com/in/mikikobazeley/