Vedanuj Goswami

@vedanujg

Research Engineer @MetaAI

Menlo Park, CA

vedanuj.github.io

Joined October 2016

Vedanuj Goswami
@vedanujg
Jul 23, 2024
Llama3 405B is out with impressive performance across the board and 128k ctx length! Along with improved models for 8B and 70B. Blog : llama.meta.com/llama-download… Paper : ai.meta.com/research/publi… A short 🧵 on what went into scaling and training the largest compute Llama model yet
Vedanuj Goswami
@vedanujg
Apr 24, 2024
It is really amazing how good these models turned out to be!
Ivan Fioravanti ᯅ
@ivanfioravanti
Apr 23, 2024
Look at this! Llama-3 70B english only is now at 1st 🥇 place with GPT 4 turbo on @lmsysorg Chatbot Arena Leaderboard🔝 I did some rounds too and both 8B and 70B were always the best models for me. Incredible achievement @AIatMeta
Vedanuj Goswami
@vedanujg
Apr 22, 2024
In addition to data and model optimization, stability, efficiency and fault tolerance of the entire infra stack is extremely crucial when we scale LLM training to tens of thousands of GPUs. Really glad to be working with @reducescatter and the brilliant AI Infra team @AIatMeta,
Jenya
@reducescatter
Apr 18, 2024
@karpathy's insights on the complexities of training LLMs really hit home. Keeping the #llama3 training alive was a journey filled with hard challenges across the entire tech stack. Reading it kept me sane, knowing others have a similar experience.
Vedanuj Goswami
@vedanujg
Apr 18, 2024
Happy to be part of this incredible journey of Llama3 and to share the best open weight 8B and 70B models! Our largest 400B+ model is still cooking but we are providing a sneak peek into how it is trending! Check more details here
ai.meta.com
Introducing Meta Llama 3: The most capable openly available LLM to date
Today, we’re introducing Meta Llama 3, the next generation of our state-of-the-art open source large language model. In the coming months, we expect to share new capabilities, additional model sizes,...
Vedanuj Goswami
@vedanujg
Jul 19, 2023
Excited to be part of the Llama 2 release today and congrats to the entire team for this massive effort! A short thread on my thoughts on the pretrained models, why the results are interesting and where there is room for improvement. 🧵 (1/n) ai.meta.com/llama