the trend is only one way – one of MIT Lincoln Laboratory Supercomputing Center (LLSC) techniques can reduce the energy of training AI models by 80%.
COMPUTING
-
Repetition in GPT-3 prompts elicits base-model-like generations, now confirmed
By
–
Fascinating work — I noticed repetition elicits base-model-like generations in GPT-3 last year but assumed they were fully hallucinated. Kicking myself now. As of August you could also do token repeats within a prompt to get base-like completion of grammatically incomplete text:
-
Python Auditory Toolbox: Audio Processing for AI Applications
By
–
MalcolmSlaney/python_auditory_toolbox: This is a Python implementation of the Auditory Toolbox https://
bit.ly/410pdOT
#AI #MachineLearning #DeepLearning #LLMs #DataScience -
Amazon Trains Next-Gen Titan LLM with NVIDIA NeMo
By
–
Explore how @amazon leveraged the NVIDIA NeMo framework, GPUs, and EFA from @awscloud to train its next-generation LLM, giving some of the largest Amazon Titan foundation models customers a faster, more accessible solution for #generativeAI. #AWSreinvent
-
GPU Memory Limits for Large Model Training on A100
By
–
No I’ve gone higher, 2k would OOM error on an 80gb A100
-
Untether AI Partners UCI Express for Chiplet AI Acceleration
By
–
@UntetherAI has joined the @UCIexpress to propel chiplet technology & energy-centric AI acceleration. UAI will enable the development of compact + high-performance compute systems across domains ranging from high-performance computing to edge applications. https://
tinyurl.com/4c8tzf96 -
Best LLM Explanation: How Large Language Models Work
By
–
I've watched the whole thing twice now (myself – No AI). It is probably the best explanation I've seen so far on how LLMs work.
-
AWS NVIDIA Partner on Generative AI Supercomputing Infrastructure
By
–
We announced a collaboration with @AWS to offer new supercomputing infrastructure, software and services for #generativeAI. This includes AWS to offer first cloud AI supercomputer with NVIDIA Grace Hopper Superchip and more.
-
Orange trains an AI to predict network coverage in seconds
By
–
Les équipes d’Orange ont construit une base de données à partir des données cartographiques , des antennes et de la couverture renvoyées par chacun de nos téléphones pour entraîner une IA qui va prédire la couverture réseau en quelques secondes pour 1 km carré ! La simulation de… pic.twitter.com/WZp1G94J0M
— Defend Intelligence (Anis Ayari) (@DFintelligence) 28 novembre 2023Orange teams built a database from cartographic data, antennas, and coverage returned by each of our phones to train an AI that will predict network coverage in seconds for 1 square kilometer! The simulation of
-
Creation of a satellite image database for diseased trees
By
–
Thus, one can build a database by labeling satellite images of diseased trees on agricultural land to enable a large-scale automatic detection and prevention solution to help small farmers.