-
Microsoft Researchers Propose PIT (Permutation Invariant Transformation): A Deep Learning Compiler for Dynamic Sparsity
Researchers at Microsoft have proposed a deep learning compiler called Permutation Invariant Transformation (PIT) to optimize models for dynamic sparsity. PIT leverages a mathematically proven property to consolidate sparsely located micro-tiles into dense tiles without changing computation results. The solution accelerates dynamic sparsity computation by up to 5.9 times compared to state-of-the-art compilers and offers…
-
McMaster University and FAIR Meta Researchers Propose a Novel Machine Learning Approach by Parameterizing the Electronic Density with a Normalizing Flow Ansatz
Researchers from McMaster University and FAIR Meta have developed a new machine learning technique called orbital-free density functional theory (OF-DFT) for accurately replicating electronic density in chemical systems. The method utilizes a normalizing flow ansatz to optimize the total energy function and solve complex problems. This approach shows promise for accurately describing electronic density and…
-
‘Lookahead Decoding’: A Parallel Decoding Algorithm to Accelerate LLM Inference
Lookahead decoding is a novel technique that improves the speed and efficiency of autoregressive decoding in large language models (LLMs) like GPT-4 and LLaMA. It eliminates the need for preliminary models and reduces the number of decoding steps by utilizing parallel processing. The technique has been shown to significantly decrease latency in LLM applications like…
-
ETH Zurich Researchers Introduce UltraFastBERT: A BERT Variant that Uses 0.3% of its Neurons during Inference while Performing on Par with Similar BERT Models
UltraFastBERT, developed by researchers at ETH Zurich, is a modified version of BERT that achieves efficient language modeling with only 0.3% of its neurons during inference. The model utilizes fast feedforward networks (FFFs) and achieves significant speedups, with CPU and PyTorch implementations yielding 78x and 40x speedups respectively. The study suggests further acceleration through hybrid…
-
Introducing three new NVIDIA GPU-based Amazon EC2 instances
Amazon announces the expansion of its EC2 accelerated computing portfolio with three new instances powered by NVIDIA GPUs: P5e instances with H200 GPUs, G6 instances with L4 GPUs, and G6e instances with L40S GPUs. These instances provide powerful infrastructure for AI/ML, graphics, and HPC workloads, along with managed services like Amazon Bedrock, SageMaker, and Elastic…
-
New method uses crowdsourced feedback to help train robots
A novel technique allows an AI agent to use data crowdsourced from nonexpert human users to learn and complete tasks through reinforcement learning. This approach trains the robot more efficiently and effectively compared to other methods.
-
AI-generated sexually explicit material is spreading in schools
Children in the UK are using AI image generators to create indecent images of other children, according to the UK Safer Internet Centre (UKSIC). The charity has highlighted the need for immediate action to prevent the problem from spreading. The creation, possession, and distribution of such images is illegal in the UK, regardless of whether…
-
“Authentic” the Merriam-Webster word of the year, but why?
Merriam-Webster has chosen “authentic” as its Word of the Year for 2023 due to its increased relevance in the face of fake content and deep fakes. The word has multiple meanings, including being genuine and conforming to fact. This decision reflects the current crisis of authenticity in a world where trust is challenged by the…
-
Boost inference performance for LLMs with new Amazon SageMaker containers
Amazon SageMaker has released a new version (0.25.0) of Large Model Inference (LMI) Deep Learning Containers (DLCs) with support for NVIDIA’s TensorRT-LLM Library. This upgrade provides improved performance and efficiency for large language models (LLMs) on SageMaker. The new LMI DLCs offer features such as continuous batching support, efficient inference collective operations, and quantization techniques.…
-
Unveiling the Frontiers of Scientific Discovery with GPT-4: A Comprehensive Evaluation Across Multiple Disciplines for Large Language Models
Language models like GPT-4, which are part of the field of Artificial Intelligence, have gained popularity due to their remarkable capabilities in various fields. These models excel in tasks such as coding, mathematics, law, and understanding human intentions. GPT-4 can process text, images, and even display characteristics of Artificial General Intelligence (AGI). Recent research has…