Code @ Anthropic. Previously Amazon AGI Autonomy, @AdeptAILabs. Ph.D. @Berkeley_EECS. Undergrad @IITKanpur.
Indianapolis, IN
Joined April 2010
- Tweets73
- Following201
- Followers611
- Likes937
Rohan Bavishi retweeted
A statement on the comments from Secretary of War Pete Hegseth.
anthropic.com/news/statement…
Bonus: We DPO'ed Fuyu-Heavy on the public Ultrafeedback dataset and got competitive results on MT-Bench and AlpacaEval 1.0! Plus it powers the model used in the chat demo you see here, underlining the balance between image and language modeling that Fuyu-Heavy achieves.
This quoted post is unavailable.
Rohan Bavishi retweeted
From spreadsheets to Airbnb listings, the ability to 'zoom out' using AI lets us quickly understand and act. It's not just about more data; it's about clearer, concise, actionable insights.
A look into what zooming out could look like from @Wattenberger and @AdeptAILabs
Rohan Bavishi retweeted
I’ve been working with @AdeptAILabs and we’ve made FlashAttention even faster for long sequences! For seqlen 8K, FlashAttention is now up to 2.7x faster than a standard PyTorch implementation even at small batch, making it easier to train better LMs with longer context 1/7
Announcing FlashAttention, a fast and memory-efficient attention algorithm with no approximation! 📣 w/ @realDanFu
By reducing GPU memory reads/writes, FlashAttention runs 2-4x faster & requires 5-20x less memory than PyTorch standard attention, & scales to seq. length 64K. 1/
Rohan Bavishi retweeted
This might have gotten lost, but this vid demonstrates voice input!
Complaints to me if you don’t like the voice in question :)
This quoted post is unavailable.
Rohan Bavishi retweeted
Train BERT on Smartphones 🤩 Announcing POET 📢 Find out how we train memory-hungry SOTA models on smartphones! #ICML2022 Thu 21 Jul 3:45 pm EDT at Room 327 🧵👇
Paper: arxiv.org/abs/2207.07697
Web: poet.cs.berkeley.edu/