Desi(@desicochrane) 's Twitter Profileg
Desi

@desicochrane

πŸ— Building @verydiscoapp. πŸ§‘β€πŸŽ“ Studying at @UniofOxford.
πŸ“š Currently reading: Faust

ID:97845458

linkhttps://blog.des.io calendar_today19-12-2009 07:07:49

532 Tweets

339 Followers

1,4K Following

Desi(@desicochrane) 's Twitter Profile Photo

At the time, I wasn't a fan of the repetitive and heavy-handed essay Tolstoy interwove War and Peace. But I confess, I likely think about it more as a result.

account_circle
Desi(@desicochrane) 's Twitter Profile Photo

Watching 3BP English adaptation with the gf is agony. She was not impressed at the onset and has since actively, and loudly, sought any excuse to further hate on it.

account_circle
Desi(@desicochrane) 's Twitter Profile Photo

From tears of pain to tears of joy. I haven't benchmarked, but running this 4090 the last couple days has been a massive productivity boost of paperspace A6000. I'm getting done in a couple of hours what I failed to get done over 2 weeks. I might get another.

account_circle
Desi(@desicochrane) 's Twitter Profile Photo

The state of pytorch + cuda + OS version compatibility is absurd. Looks like Debian 12 won't work and I'll either need to downgrade to Debian 11 or switch to Ubuntu 22.04. Feels wrong. Also, will I regret going with cuda 12.1? Does spaCy support cuda 12x? Burning cycles here.

account_circle
Desi(@desicochrane) 's Twitter Profile Photo

I'm having to fill in some sysadmin gaps in my resume to set up this home rig. Finally got the storage setup how I want: Going with RAID5 across three 4TB SSDs, encrypted at rest, with LVM on top. It was very difficult to find good documentation for any of this. Next stop: cuda

account_circle
Desi(@desicochrane) 's Twitter Profile Photo

Anyone have issues with Paperspace gradient kernels just failing from time to time during memory intensive work? Super annoying, the task I'm working on takes over 20 min to load the stuff I need into memory, so every kernel restart is a massive crush to productivity.

account_circle
Desi(@desicochrane) 's Twitter Profile Photo

Today I'm migrating some 2020 pytorch_transformer models to latest so I can fine-tune. I'm already having trouble with the state_dict not lining up. No tests either, and its memory heavy so it keeps crashing my paperspace instance. Still, the results are looking promising.

account_circle
Tsarathustra(@tsarnick) 's Twitter Profile Photo

Nick Bostrom asks: what if AI could form such an accurate model of you that it could tell what you are thinking and persuade you to do anything?

account_circle
Desi(@desicochrane) 's Twitter Profile Photo

I'm itching to get to the fun stuff, but I have a few broken windows to repair first. This morning: auth. 🀒

account_circle
Desi(@desicochrane) 's Twitter Profile Photo

I have a serious problem. I forgot to bring my current read (Faust) to Oxford this week, and now I'm returning home with 12 new books wtf Waterstones must be stopped.

I have a serious problem. I forgot to bring my current read (Faust) to Oxford this week, and now I'm returning home with 12 new books wtf @Waterstones must be stopped.
account_circle
Desi(@desicochrane) 's Twitter Profile Photo

Last week I was rewatching what is hands-down *the* best explanation of 'attention' online (via the Rasa YT channel) when I recognize the voice as belonging to none other than the legend Vincent D. Warmerdam from the spacy gang! What a friggin champion! Love your work mate πŸ™‡β€β™‚οΈπŸ™‡β€β™‚οΈπŸ™‡β€β™‚οΈ

account_circle