Henok | Neural Nets

ጸረ AI እንቅስቃሴ at Mila😁 It'd make sense if this happened in SF than here. https://www.cbc.ca/news/canada/montreal/mila-vandalism-montreal-ai-summit-9.7345851

Happy new year 🌼

Hello, for anyone currently working on thesis or projects in AI fields. Please fill out this form, we are trying to create a good support system and mentors/advisors. https://docs.google.com/forms/d/e/1FAIpQLSdJYcasoScmhkfOJt-pT8UIdfaOkwrgIDPUte2qZeARu-Pnhg/viewform?usp=publish-editor

I wish am a part of some univ in eurpe, canada or usa for real people can be creative and test many things. as per our country damn guys we need the resource sooooo much to even be creative enough

Unpopular opinion, when ever I see a crying baby it's usually the black ones. Compared to Ethiopia and the white kids here I don't usually see/hear them crying. It's weird. Btw anyone tried Astra 😁

One of the worst parts about doing research, whether you’re developing models or building benchmarks, is that these labs keep releasing a new model every two weeks 😭 You can spend months running experiments, finally get everything ready to publish, and then suddenly a new model drops that’s better than everything you evaluated. Now you’re sitting there like…, time to rerun, train again (hopefully not the pretraining part at least) , rethink the conclusions, so boringgg

Went to a Brazilian Jiu-Jitsu class and jumped in with the advanced guys. Learned a few moves and actually really liked it, but those sparring rounds are no joke

Can't wait to see the step from GPT to Astra or is it going to be GPT Astra. AGI though, I don't think so but depends on their definition which is "highly autonomous systems that outperform humans at most economically valuable work". There's still a lot to do in AI and explore.

We're releasing the first batch of Dataset.ET open speech data for Amharic. 22.7 hours. 7,405 recordings. 320 speakers. Free, CC BY 4.0, on Hugging Face. Amharic has close to no open speech data. That's the gap we're trying to close, and this is the first step rather than the finished thing. Some honesty about what this is: it's a first batch. The audio isn't preprocessed yet. We're not going to tell you it's the best quality out there. We don't think it is. What we want is for people to actually use it and tell us where it falls short. The next release will be different: properly preprocessed audio, a real data pipeline with testing and validation built in. This exists because 320 people in Ethiopia gave their voices to it, and because contributors reviewed each other's recordings. Thank you to all of them. If you're working on Ethiopian languages, low-resource ASR, or you just want to talk about the data, DM us. We'd genuinely like to hear from you. 🔗 https://huggingface.co/datasets/snapwre/amharic-speech

Episode 5 is finally here! 🎙️🔥 This time, I sat down with Mohammed Ibrahim a.k.a Doughnut. We talked about ExamBuddy, building AI products from Ethiopia, remote opportunities and the future of software development. this one is packed with knowledge, honest conversations and plenty of laughs 😂 #SoftwareGuyPodcast #EthiopianTech #AI #Programming https://youtu.be/tm3y6rnYyWY

really "AI መስራት ቀላል ነው" ነው Doughnut😁? Seems like a fun podcast though👏

And in general this is the goto book for RL might feel outdated if you want fake RL/LLM RL, the ones everyone was doing the past couple of years for LLMs but now real RL is being used a lot.

Also here is a great slides on LLM Post Training https://kawine.github.io/assets/aiesi_post-training_public.pdf

I've been focusing on LLM post training lately and whether you're doing very simple fine-tuning or any kind of SFT or wanted to do some RL on top of SFT models using online RL like GRPO, just use LlamaFactory, I've been skeptical about them at first, but they are increasingly becoming very common and it makes things very easy to work with. https://github.com/hiyouga/LlamaFactory

Maybe this is why Jelani left his academic job 😂. But wait Jelani also knows about Robi's SMA 😂? But I am proud to say I've worked with Shalom during my time at bete semay while working on moogter and I have to say bro is a rare breed of an engineer.

Wait so Man United lost against Hull City was a premiere league match? Who was going to tell me, tf is this damnnnn, here we go again

1. Currently i'm on a break from my job focusing on research. It involves reading, running experiments, analyzing results, talk with my advisor, not daily but 2. This is nice a blog i usually send it to people. I just like ML and that's all, and still studying so not smart enough yet, there's more to do and learn

who the hell are you😂 life after mila is almost the same i'd say but more proper research, more reading, more focused, at times more boring too I don't think people like me or hate me, so i'd say wrong assumptions hehe

I'm friends with many but if we say close then it's chapi, meron, simon and luna, unless they deny this lol

Life isn't bad, dating i don't know bruh