From my personal experience, it's best to find a windows laptop that supports two m2 nvme slots. Get another cheap ssd and install linux exclusively on it. I have this setup on my laptop and didn't have to deal with wiping anything.
I'm a grad student and researcher at UC Merced. My academic focus and hands-on experience is in training ML/DL models and optimizing inference performance. Professionally I have ~3 yrs of exp as a Backend Software Dev developing payment processing pipelines and integrating them with payment settlement systems.
I am looking for opportunities as an ML/DL engineer or Backend SDE.
Absolutely horrible! I am about to complete my MS EECS degree and have been applying to SWE or MLE roles. Even with my previous ~3 years of Backend Dev experience I am not able to get past resume screening. I just get auto rejected all the time.
Doesn't help the fact that I have been in the US for a little bit more than a year and half, so building a network is pretty hard. Attended a hackathon, won 2nd prize with my team and tried to leverage something from there but no luck at all.
Makes me feel that I might be missing something but no idea what's wrong with my profile. (I don't need sponsorship too).
Technologies:Python,FastApi,Flask, Java, Spring Framework, PyTorch, Tensorflow, C/C++, PostgreSQL, MATLAB,ROS. A little bit of JS with React and express.js.
I'm currently an MS EECS student and a Graduate Student Researcher at UC Merced. My academic focus and hands-on experience are in machine learning, deep learning, and computer vision. I focus on optimizing Deep Learning model performance.
Professionally I have ~3 years of exp as a Backend Software Dev developing payment processing pipelines and integrated them with third-party payment settlement systems (HKICL, BPay to name a few).
I am looking for opportunities as an ML/DL engineer or Backend SWE once I graduate (May 2025).
Technologies:Python,FastApi,Flask, Java, Spring Framework, PyTorch, Tensorflow, C/C++, PostgreSQL, MATLAB,ROS. A little bit of JS with React and express.js.
I'm currently an MS EECS student and a Graduate Student Researcher at UC Merced. My academic focus and hands-on experience are in machine learning, deep learning, and computer vision. I focus on optimizing Deep Learning model performance.
Professionally I have ~3 years of exp as a Backend Software Dev developing payment processing pipelines and integrated them with third-party payment settlement systems (HKICL, BPay to name a few).
I am passionate about solving complex problems and delivering high-impact solutions. I am looking for opportunities as an ML/DL engineer or Backend SWE once I graduate (May 2025).
Column looks very good. The API first approach is something that has been absent from banks and financial institutes for a long time now. I have worked with a lot of established banks trying to integrate their payment systems with newer tech. I would love to work on stuff that Column is doing. Have sent an email to you Wiiliam.
Hi Mark, the library looks cool, excited to try it out. Coincidentally I am starting work on a project that is investigating a lot of Post training quantization methods. I read the blog and I am curious to understand what kind of overheads are involved in quantizing a layer?
There's a bunch of overhead associated with PTQ - but TL;DR is that much of that overhead goes away when you're using `torch.compile()` and `torchao.autoquant()`
Essentially the latency overhead comes from quantizing and dequantizing weights and activations. For large layers this overhead is small because by quantizing your weights for example you reduce memory bandwidth pressure but for small layers the overhead of potentially looking up a table, reading scaling factors, quantization/dequantization and finally handling zero points might not be worth it.
However, even if such overhead exists you can still quantize your model and get it to be smaller it might not be faster is the problem. We solve the speed problem in 2 ways - `torch.compile()` will fuse operations like a dequant and matmul into a single kernel and `torchao.autoquant()` will do kernel level profiling to see whether a layer is actually made faster when quantizing and if not it skips quantizing that layer.
