Was chatting with
@marmars
this morning and discovered we are both into jazz! Think might be cool to drop some improvisation (ie unnecessarily long) videos here
@X
Just watched this lecture and built a GPT, super interesting
- encode/decode characters into indexes and vice versa
- bigram language model as the baseline
- a generate function to make predictions given some contexts via softmax mapping and concat sequences
- self-attention:
Vector similarity search allows us to search a huge range of media for billion+ size datasets. Some of the most important indexes including Flat, LSH, HNSW, and IVF. Some factors to decide which indexes including Dataset size, Search frequency, or Search-quality vs. Search-speed
hi! appreciate a ⭐ for this Github open source project! My brother
@hsu_byron
, is the 1st author
Introducing Liger-Kernel: Efficient Triton Kernels for LLM Training
Add one line to boost multi-GPU training throughput by 20% and reduce memory usage by 60%.
(1/n)
Training LLMs can be hindered by out-of-memory, scaling batch size, and seq length. Add one line to boost multi-GPU training throughput by 20% and reduce memory usage by 60%. Introducing Liger-Kernel: Efficient Triton Kernels for LLM Training.
Hi everyone! We are hiring ML engineering interns for this summer for both our product ranking and ML platform teams!! Please share with anyone who might be interested in joining us!!
“even if ML can’t solve your problem, it might be possible to break your problem into smaller components, and use ML to solve some of them. For example, if you can’t build a chatbot to answer all your customers’ queries, it might be possible to build an ML model to predict
And today marks my last day. It has definitely been quite a ride. I learned and grew a lot, and got to work with many incredible people. Time for a little break.
“Tips for training data preparation : 1. different sampling techniques 2. challenges in creating training data - label multiplicity, the lack of labels, the class imbalance 3. Techniques in data augmentation - lack of data”
“Data in production is neither finite nor stationary,
This has been a year full of engineering excellence that sometimes can go unnoticed. Besides all the visible changes you see on our app, here are some of the most important improvements we have made under the hood.
- Consolidated the tech stacks for For you, Following, Search,
Today marks a new era of transparency for Twitter. 🧵
We’re sharing much of the source code that powers our platform with the world. Visit our blog to learn more about this initiative:
Hi friends, I am hiring data engineers here at X. Please reach out if you’d like to solve hard and interesting problems (AI/ML related) and willing to relocate to SF office, experience with Spark/Flink/Scalding/Kafka is a plus, DM is open!
7. “Good Will Hunting” (1997) - Farting Wife Story
The scene features a monologue by Robin Williams, who played the character of Sean Maguire, a therapist. In this scene, he shares a humorous anecdote about his late wife’s habit of farting in her sleep. This dialogue was not
“Feature engineering includes handling missing values, scaling, discretization, encoding categorical features, generate old-school but effective cross features, and newer and exciting positional features”
3 types of missing values, and 2 ways to address them - Missing not at
Excited to start sending our first payments to creators for ads revenue sharing today. Creators are the lifeblood of this platform, and it's great to see so many creators I follow getting paid today. The program will be expanding soon—more to come!
Hello!! I'm hiring ML Infra Engineers at X. Please reach out if you know anyone who's interested in further building out infra layer to support cutting edge AI/ ML Products!
“Data leakage refers to a mistake where information from the test set inadvertently influences or contaminates the training process. Basically cheated by having access to some information it shouldn’t have had during training” - ChatGPT
“For instance, imagine you’re trying to
Foundations of HNSW - We can split ANN into 3 distinct categories; Trees, Graphs (HNSW) and Hashes. HNSW is a proximity graph, in which two vertices are linked based on their proximity (closer vertices are linked),often defined in L2. Two fundamental techniques that bulit HNSW:
Check out the community I created:
For jazz artists: post your upcoming shows & gigs here!
For jazz venues/organizers: share your upcoming events!
For jazz cats: enjoy the shows!
Starting today, we're testing a new program (Not A Bot) in New Zealand and the Philippines. New, unverified accounts will be required to sign up for a $1 annual subscription to be able to post & interact with other posts. Within this test, existing users are not affected.
This
Today is the day: Ads Revenue Sharing is now live for eligible creators globally.
Set up payouts from within Monetization to get paid for posting.
We want X to be the best place on the internet to earn a living as a creator and this is our first step in rewarding you for your
Last week, I flew back to Taiwan due to family reasons and began my 14-day self isolation. As I am close to the end of the journey, I want to share my experience and provide a glimpse of how this small island manages to effectively contain COVID and prevent local transmission.
We've been busy this week making several improvements to Communities. Here are the most notable changes that are coming soon:
- We're updating Pinned Communities on Home to have sorting options: Trending, Most liked, Most recent.
- We are also adding a small version of the
One of my fav jazz guitarist, Julian Lage, is in town! it is a celebration of the late guitar legend Jim Hall. I am going. Anyone?
1/21 Sunday in SF Jazz
Many people ask how to become a creator, tap [Professional Tools] -> [Monetization], apply today! Enjoy and let us know your feedback!
We will also make it easier to find.