“a billion parameters of local data”? 👀 Brooo, this phrasing doesn’t even make sense technically.
Secondly, 78% “accuracy” on local dialects - what metric? WER? Custom test set? Without details or reproducible evals, this just sounds like classic hype.
“train, re-train, extra training, during training, 'training the model in it's initial stages' …”
Training even a 1B model from scratch requires way more than a single GPU - distributed training, for something like TinyLlama 1.1B took about 16 A100-40G GPUs for a span of 90 days. I hope he meant fine-tuning, which I also highly doubt.
"42 Languages" 🙄 ... No technical details provided - no architecture, no data curation method, no training process, compute used or reproducible evals.
Did you also “train the model” to write Rust and Golang? Convince me this isn’t just a GPT wrapper.
I support the effort, but some things need to be called out for clarity.
Today has been a difficult day in Ukraine after a massive Russian strike. There were many ballistic missiles and drones. Now air-raid alerts are active across various regions too. And once again, we are facing aerial threats. Just today, Russian strikes have taken 22 lives in Kyiv and Dnipro, including children. We are doing everything to protect our people, our cities, and our communities. And we are grateful to those who are helping us. Ukraine is grateful to Italy. We deeply value that you care, that you want peace for us, and see the defense of Kyiv’s independence and freedom as your cause. Together, we will certainly achieve this.
I congratulate Italy, President Mattarella @Quirinale, President of the Council of Ministers @GiorgiaMeloni, and all Italian people on the 80th anniversary of the proclamation of the Republic. Freedom is an achievement that must be defended every day.
Thank you, Italy! Buona Festa della Repubblica!
To the black KDV Mazda CX-5 this morning...Usikubali kupewa stress na Corolla 91. I know, your car feels fast but a random Corolla isn't the one to try bullying 😎. Round 2?
The lead maintainer for linux kernel memory subsystem is stepping away, hehas been doing this job for 26 years. So far they haven't figured out a clear path to replacement & considering this subsystem accounted for 17% of linux kernel CVE's between 2020-2024,are we fucked?