Excited to share insights from our first-ever multilingual models workshop at NeurIPS 2023! We are making available video recordings from each expert who joined us to discuss multilingual model development. Highlights include:
– Addressing dataset challenges for
RESEARCH
-
NeurIPS 2023 Multilingual Models Workshop: Expert Insights
By
–
-
YOLO Runs: Training Neural Networks Explained
By
–
How I call "yolo runs": training a neural network.
-
The Art of YOLO Runs in AI Research at OpenAI
By
–
An incredible skill that I have witnessed, especially at OpenAI, is the ability to make “yolo runs” work. The traditional advice in academic research is, “change one thing at a time.” This approach forces you to understand the effect of each component in your model, and
-
Why AI Still Struggles with 3D Rigging Tasks
By
–
I'm actually a bit surprised that AI isn't better at rigging yet.
-
Canada AI Future Summit at Empire Club This Thursday
By
–
We're looking forward to an in-depth conversation on Canada's #AI future, this Thursday at the @Empire_Club
. Seats are filling up rapidly for this one-of-a-kind summit, bringing together expertise and inquiry to map AI's possibilities. Register now! http://
hubs.la/Q02fTH-D0 -
Stable Cascade Improves Hand Generation in Image Synthesis
By
–
The only benchmark you've been waiting for:
— Pietro Schirano (@skirano) 13 février 2024
Balenciaga Pope on Stable Cascade.
Hands have improved so much! WOW! 🙌 https://t.co/hYpbrEMDcH pic.twitter.com/4dpcik7ShEThe only benchmark you've been waiting for: Balenciaga Pope on Stable Cascade. Hands have improved so much! WOW!
-
Lumiere: Space-Time Diffusion Model Generates Coherent Videos
By
–
Lumiere is a space-time diffusion research model that generates video from various inputs, including image-to-video. The model generates videos that start with the desired first frame & exhibit intricate coherent motion across the entire video duration → https://t.co/QAMgC4TmBL pic.twitter.com/CZCDDfpMAJ
— Google AI (@GoogleAI) 13 février 2024Lumiere is a space-time diffusion research model that generates video from various inputs, including image-to-video. The model generates videos that start with the desired first frame & exhibit intricate coherent motion across the entire video duration → https://
goo.gle/47WX6C2 -
Getting Started with LLMs: Introduction, Quantization, and Evaluation
By
–
🚀 Blog-tastic Tuesdays 🚀https://t.co/ThNcwPHhyv
— Satya Mallick (@LearnOpenCV) 13 février 2024
Getting started with LLMs can be overwhelming. With our new article, we jump-start a series on LLMs, and the first one ventures into the introduction, quantization techniques, and evaluation metrics for LLMs.#computervision… pic.twitter.com/wb2EzQPNmiBlog-tastic Tuesdays https://
learnopencv.com/deciphering-ll
ms/
… Getting started with LLMs can be overwhelming. With our new article, we jump-start a series on LLMs, and the first one ventures into the introduction, quantization techniques, and evaluation metrics for LLMs. #computervision -
LLMCompiler: Streaming Tasks as DAG for Immediate Execution
By
–
3/ LLMCompiler
Stream tasks as a DAG. Immediately execute once dependencies are resolved, so you can invoke functions while the planner LLM is still generating output. Based on research of @sehoonkim418 , et. al Python: https://
github.com/langchain-ai/l
anggraph/blob/main/examples/llm-compiler/LLMCompiler.ipynb
… -
ReWOO: Task Output Assignment for Focused Agent Execution
By
–
2/ ReWOO Assign task outputs to variables that can be explicitly passed to subsequent tasks. Provides more focused context to task executors and makes it easy to fine-tune a planner LLM (since the plan doesn't depend on tool outputs) Based on research of @billxbf
, et. al.