Alexander Morgan (@lexandermorgan) header
Alexander Morgan (@lexandermorgan) profile picture

Developer at Datamaran. Tinkering with tokenization, hypermedia, and sustainability.

Barcelona, España
Joined August 2019
Photo by Alexander Morgan (@lexandermorgan)Photo by Alexander Morgan (@lexandermorgan)Photo by Alexander Morgan (@lexandermorgan)Photo by Alexander Morgan (@lexandermorgan)Photo by Alexander Morgan (@lexandermorgan)Photo by Alexander Morgan (@lexandermorgan)Photo by Alexander Morgan (@lexandermorgan)Photo by Alexander Morgan (@lexandermorgan)Photo by Alexander Morgan (@lexandermorgan)Photo by Alexander Morgan (@lexandermorgan)
Hypermedia practitioners unite!
Calling all employers — If you are hiring for a stack with a big back end (you know who you are), post your job with @lexandermorgan new site Hypermediajobs.com As I talk about in the episode linked below: there are more of these jobs than it might appear!
1
17
Definitely not for everyone. @htmx_org is html for gen-Xers. It's in the name!
Replying to @JLarky
Okay, I did the call, the guy was so confused about who exactly the target audience is for htmx, but I tried to convince him that it’s everyone!
1
1
10
2,209
Hey @RhysSullivan when adding an integration to Executor, is it better to go with the mcp or the cli if they're both available eg for github?
1
785
I'm sad to report that nobody plays #eightile in Galician, and that Spanish and French aren't much better.
25
Designing a new bathroom with @tldraw and I noticed the counter-clockwise keyboard shortcut doesn't display correctly. It's shift + , but it just shows up as shift. tldraw is definitely worth a try 👌
1
3
551
With memory prices skyrocketing, the best quick fix is for software to become more mem efficient. Time to dig in to mem optimizations!
1
24
Palomitas is done! 🍿 It's a game where you're a pigeon scavenging for food. Part of #vibejam 2026. palomitas.pages.dev
2
81
Palomitas is now up and running for #vibejam You're pigeon that tries to steal food. Cool idea but not really fun yet. Lots of work to do still.
3
3
71
I just spent ~90min vibe-planning my entry for #vibejam How do other people go so much faster? If you're not actively participating in at least the plan I don't get how you learn anything from prompting.
32
Open letter to @NYCMayor asking for the removal of NYC's harmful trash subsidy. If you're wondering how cities subsidize trash creation and littering, give it a read.
1
31
If you need a hand with this, feel free to reach out.
Replying to @DanielLockyer
I need to scan Coolify as well for more indexes, because the cloud version is a bit busy. 😅
24
Replying to @classiclarryd
@classiclarryd I can never get H100s on PrimeIntellect with pytorch pre-installed so the docker builds don't work. With Ubuntu I get slower nanogpt runs (88s) without changes, but 85.6s on my branch with changes. Does this mean mine would be a record if it ran with docker?
1
42
I'd raise a pr but my branch is suspiciously the same exact runtime as the record 🧐 github.com/alexandermorgan/m…
1
29
Hey @marclou, I'm trying to post a website for sale on TrustMRR. It doesn't have any revenue but I am still required to submit a Stripe key for it. I see that there are other zero-revenue listings, is there some way around the Stripe requirement in this case?
1
40
Last month at PyConES I shared my approach to safely batching BPE #tokenization merges. This boils down to: 1. Deciding whether a pair is safe to merge 2. Determining how many pairs to consider in a single batch
1
1
2
130
Together these 2 points mean the freq. of any token pair, eg "th", can only decrease as it's fragmented by other merges, eg "_t" or "he". So later there can't be other pairs with more than whatever pair you're considering merging since neither of its tokens have been seen.
1
56
Batching tokenization merges is the main optimization I integrated into BatchBPE, my fork of @karpathy's minbpe. It gives a ~300X speedup! It would be great to see this approach used in a tokenizer written in a different language. github.com/alexandermorgan/B…
52