Stability AI's Code Assistant Challenges Meta's Code Llama 7B
Plus: Alibaba announces AI Motionshop, ArtificialAnalysis guide you toward best LLM.
Hello Engineering Leaders and AI Enthusiasts!
Welcome to the 190th edition of The AI Edge newsletter. This edition brings you Stability AI’s New Coding Assistant to Rival Meta's Code Llama 7B.
And a huge shoutout to our incredible readers. We appreciate you😊
In today’s edition:
💻 Stability AI’s new coding assistant to rival Meta's Code Llama 7B
✨Alibaba announces Motionshop, AI replaces video characters in 3D avatars
🔍 ArtificialAnalysis guide you select the best LLM
🧠 Knowledge Nugget: Optimizing Distributed Training on Frontier for Large Language Models by
Let’s go!
Stability AI’s new coding assistant to rival Meta's Code Llama 7B
Stability AI has released Stable Code 3B, an AI model that can generate code and fill in missing sections of existing code. The model, built on Stability AI's Stable LM 3B natural language model, was trained on code repositories and technical sources, covering 18 different programming languages.
It outperforms other models in completion quality and is available for commercial use through Stability AI's membership subscription service. This release adds to Stability AI's portfolio of AI tools, including image, text, audio, and video generation.
Why does this matter?
Their ability to develop performant models with fewer parameters than competitors like Code Llama shows their technical capabilities. Providing developers access to advanced coding assistance AIs allows faster and higher quality software development. And its multi-language support also makes AI-assisted coding more accessible.
Alibaba announces Motionshop; AI replaces video characters in 3D avatars
Alibaba announces Motionshop, It allows for the replacement of characters in videos with 3D avatars. The process involves extracting the background video sequence, estimating poses, and rendering the avatar video sequence using a high-performance ray-tracing renderer.
It also includes character detection, segmentation, tracking, inpainting, animation retargeting, light estimation, rendering, and composing. The aim is to provide efficient and realistic video generation by combining various techniques and algorithms.
Why does this matter?
By combining advanced techniques like pose estimation, inpainting, and more, Motionshop enables easy conversion of real videos into avatar versions. This has many potential applications in social media, gaming, film, and advertising.
ArtificialAnalysis guide you select the best LLM
ArtificialAnalysis guide you select the best LLM for real AI use cases. It allows developers, customers, and users of AI models to see the data required to choose:
Which AI model should be used for a given task?
Which hosting provider is needed to access the model?
It provides performance benchmarking and analysis of AI models and API hosting providers. They support APIs from: OpenAI, Microsoft Azure, Together.ai, Mistral, Google, Anthropic, Amazon Bedrock, Perplexity, and Deepinfra.
If you’d like to request coverage of a model or hosting provider, you can contact them.
It shows industry-standard quality benchmarks and relies on standard sources for benchmarks, which include claims made by model creators.
Why does this matter?
ArtificialAnalysis provides an important benchmarking service in the rapidly evolving AI model landscape by systematically evaluating models on key criteria like performance and hosting requirements. This allows developers to make informed decisions in selecting the right model and provider for their needs rather than relying only on vendor claims.
Example of Comparing between models: Quality vs. Throughput
Enjoying the daily updates?
Refer your pals to subscribe to our daily newsletter and get exclusive access to 400+ game-changing AI tools.
When you use the referral link above or the “Share” button on any post, you'll get the credit for any new subscribers. All you need to do is send the link via text or email or share it on social media with friends.
Knowledge Nugget: Optimizing Distributed Training on Frontier for Large Language Models
This paper discusses the optimization of distributed training on the Frontier supercomputer for LLMs using AMD GPUs. The authors trained LLMs of various sizes, including one with 1 trillion parameters, using a combination of tensor, pipeline, and data parallelism.
They adapted the Megatron-DeepSpeed framework for AMD ROCm, converting CUDA code to HIP code and obtaining ROCm versions of necessary libraries. The experiments involved testing different parallelization strategies and hyperparameters to find the optimal setup.
This interesting work by
highlights the increasing use of AMD GPUs in training large neural networks and the competition it brings to the field.Why does this matter?
This work is an important step in that direction, opening new frontiers in advanced AI training. Demonstrates the viability of using AMD GPUs and the Frontier supercomputer to train enormous AI models with trillions of parameters. Competition between hardware vendors like AMD and Nvidia will continue driving innovation on optimized software frameworks like Megatron-DeepSpeed.
What Else Is Happening❗
🤝 Vodafone and Microsoft have signed a 10-year strategic partnership
To bring Gen AI, digital services, and the cloud to over 300M businesses and consumers across Europe and Africa. The focus will be transforming Vodafone's customer experience using Microsoft's AI and scaling Vodafone's IoT business. Also, Vodafone will invest $1.5B in cloud and AI services developed with Microsoft. (Link)
👥 OpenAI is forming a new team, ‘Collective Alignment’
The team will work on creating a system to collect and encode governance ideas from the public into OpenAI products and services. This initiative is an extension of OpenAI's public program, launched last year, which aimed to fund experiments in establishing a democratic process for determining rules for AI systems. (Link)
🎙️ Adobe introduces new AI audio editing features to its Premiere Pro software
The updates aim to streamline the editing process by automating tedious tasks such as locating tools and cleaning up poor-quality dialogue. The new features include interactive fade handles for custom audio transitions, AI audio category tagging, and redesigned clip badges for quicker application of audio effects. (Link)
🔐 Researchers have discovered a vulnerability in GPUs from AI Giants
Apple, AMD, and Qualcomm could potentially expose large amounts of data from a GPU's memory. As companies increasingly rely on GPUs for AI systems, this flaw could have serious implications for the security of AI data. While CPUs have been refined to prevent data leakage, GPUs, originally designed for graphics processing, have not received the same security measures. (Link)
🍎 Apple Learning Research team introduces AIM
It’s a collection of vision models pre-trained with an autoregressive objective. These models scale with model capacity and data quantity, and the objective function correlates with downstream task performance. A 7B parameter AIM achieves 84.0% on ImageNet-1k with a frozen trunk, showing no saturation in performance. (Link)
New to the newsletter?
The AI Edge keeps engineering leaders & AI enthusiasts like you on the cutting edge of AI. From ML to ChatGPT to generative AI and LLMs, We break down the latest AI developments and how you can apply them in your work.
Thanks for reading, and see you tomorrow. 😊