Llama 3.2 by Meta is here Lightweight models + multimodal large language models that see, reason, and understand Model Size:
•Lightweight: 1B and 3B
•Multimodal: 11B and 90B Outperforms most open and closed multimodal models on industry benchmarks! Takes Llama
RESEARCH
-
Meta Launches Llama 3.2 with Lightweight and Multimodal Models
By
–
-
Smaller AI Models Emerge as the Next Big Industry Trend
By
–
also speaks to the point (which i made in a recent story — conveniently linked below!) that even the biggest players in AI increasingly see the importance of smaller and smaller models; bigger is not always better. https://
bloomberg.com/news/articles/
2024-08-08/move-over-llms-small-ai-models-are-the-next-big-thing
… -
NVIDIA GPUs Accelerate Cancer Research at Sanger Institute
By
–
The @sangerinstitute is using NVIDIA GPUs and Parabricks to accelerate cancer research, saving up to $1M and 1,000 MWh annually. #sustainablecomputing #AIforGood
-
ORNL Advances AI for Computational Sciences and Engineering
By
–
Our team had an amazing time at the Smoky Mountains Computational Sciences and Engineering Conference last week! It was a pleasure to hear from @ORNL
's Prasanna Balaprakash as he shared how @ENERGY tackles challenges across the #sciences using advanced #AI. @jenniferglore -
LLMs Master Step-by-Step Reasoning for Complex Problem Solving
By
–
LLMs now have step-by-step reasoning! Smaller models like Llama-70B and Smaug can tackle complex problems by breaking them down. This is only the beginning – they will be able to handle even more complicated tasks with ease! pic.twitter.com/bhJ6HHA4Ak
— Abacus.AI (@abacusai) 25 septembre 2024LLMs now have step-by-step reasoning! Smaller models like Llama-70B and Smaug can tackle complex problems by breaking them down. This is only the beginning – they will be able to handle even more complicated tasks with ease!
-
NVLM Paper Reveals Dataset Quality Matters More Than Scale
By
–
The NVLM paper is outstanding. It is full of remarkable findings: (1) "dataset quality and task diversity are more important than scale", (2) positive transfer from multimodal datasets to text-only on math benchmarks, (3) model and data ablations, etc. Congrats to the authors on
-
LLM Validators Alignment Human Preferences Evaluation
By
–
Who Validates the Validators? Aligning LLM-Assisted Evaluation of LLM Outputs with Human Preferences Tune in to the Paper Club today to cover Shreya's latest hit with @eugeneyan
! link below -
Pragmatism over idealism: building realistic AI systems
By
–
"hypothetical 'perfect world' with no bad/predatory actors" – I am not trying to imagine something that is not realistic. Quite the contrary. I am always interested only in something that is pragmatic and can survive in REAL life. However, this thread is just a though
-
OmniBench: Universal Omni-Language Models Research
By
–
Hacker News for AI research papers OmniBench: Towards The Future of Universal Omni-Language Models hot right now
-
Understanding Attention: K and V matrices, masking, and -inf for softmax
By
–
This is a must-watch to understand how attention works! Great visualization, explaining:
– Why the K and V matrix, what do they represent?
– Why mask the lower left part of the KV product?
– Why apply -inf to the lower left part of the KV product before softmax rather than just