dag
@theDrewDag
Followers
684
Following
4K
Media
156
Statuses
3K
Not a data scientist, but it's the closest title to what I am. I like data, machine learning and the brain. https://t.co/d3ppxCqUNo https://t.co/2riIqO5118
Italy
Joined January 2021
I have been suffering of #migraineauras for the last year. I've been to neurologists, ophthalmologists and general physicians to no avail. Magnetic resonance angiography is negative, so is EEG. Today, I woke up with an aura. Crazy feeling. This triggered the data scientist in
1
0
4
The real magic is in Fortran and COBOL, which together are running most of banking, insurance, scientific/engineering domains.
0
0
0
Researchers at JMU have built a 300×300 nm light-emitting pixel — bright enough to match conventional OLEDs thousands of times larger. In theory, that’s enough to fit a Full HD display on just one square millimeter. The team used a nanoscale optical antenna and new insulation
uni-wuerzburg.de
Physicists at the University of Würzburg have developed extremely small pixels using optical antennas and a clever design. These can be used in compact AR glasses, for example.
0
0
2
I just solved the strangest tech problem I've ever come across. My wifi kept dropping packets, confirmed by ping. It would look something like the first image (packets dropping, then it comes back to life). After a while the connection would just stop working completely and drop
746
2K
21K
Reduce @PyTorch model training time by over 50%. See the benchmarks and learn how to implement multi-worker data loading and pinned memory in this step-by-step tutorial by @theDrewDag. https://t.co/BK1tEWL60H
towardsdatascience.com
Learn how to diagnose and resolve bottlenecks in PyTorch using the num_workers, pin_memory, and profiler parameters to maximize training performance.
1
1
7
The efficiency of your #PyTorch training loop significantly impacts training time, cost, and scalability. For faster models and better resource use, adopt these strategies: - Use a DataLoader with num_workers > 0 for parallel batch loading - Enable prefetching and pin_memory
towardsdatascience.com
Learn how to diagnose and resolve bottlenecks in PyTorch using the num_workers, pin_memory, and profiler parameters to maximize training performance.
1
2
13
Is your GPU starving? An inefficient data pipeline is often the hidden cause of slow model training, not the GPU itself. @theDrewDag's new article dives into the cascading bottleneck effect. https://t.co/BK1tEWL60H
towardsdatascience.com
Learn how to diagnose and resolve bottlenecks in PyTorch using the num_workers, pin_memory, and profiler parameters to maximize training performance.
0
2
7
I don't know how uncommon this opinion is, but I would really want #Python to enforce typing in a future version. I am still writing types everywhere, so why not bundle it up at the language level instead of giving me space to introduce errors?
0
0
2
This is probably the most evil thing you'll read about in 2025. Recap: - A streamer under cancer treatment raises about 32K worth of Solana to tackle medical expenses - Gets suggested to download a game via Steam called "Block Blasters" from a random individual in chat -
decrypt.co
Raivo “Rastaland” Plavnieks is battling a rare form of cancer, and Solana funds raised for his treatment were swiped via a malware-infected game on Steam.
0
0
1
Working with @tiangolo's SQLModel is a joy few programming tools offer. Creating a table from a Pydantic model with just `table=True` is just incredible and offers great flexibility. Using data models as database structures eliminates code duplication (src/db/models vs.
1
1
8
28/08/25 #Etna Update! The activity is fluctuating but with three more or less active lava flows and strong strombolian activity at one of the summit craters - the Southeast Crater - life gets never boring on this beautiful piece of land 🌋
2
26
80
0
1
3
The next step I believe will be a game changer in software like @cursor_ai and @windsurf is allowing the LLM to use the debugger and inspect local and global variables at each debugging step. This would be very powerful and remove the errors / long conversations regarding
0
0
1
The clearest Image of Venus's Surface, by a lander that melted after 1 hour. Venus is the hottest planet in our solar system, with an average surface temperature of about 462°C/864°F, hot enough to melt lead. This extreme heat is not due to its proximity to the Sun alone
0
0
1
lmao this is very entertaining
0
0
0