- Weekly Cloud Info
- Posts
- Weekly Cloud Info #W51 - 2024
Weekly Cloud Info #W51 - 2024
Walid LARABI
December 20, 2024
Hi!
Even with the holidays in sight, we still have news from tech players. GitHub offers free Copilot access, AWS unveils powerful EC2 instances and Stable Diffusion models, and Microsoft accelerates a passwordless future. Google advances AI video generation and enterprise tools, while Hugging Face elevates language models and materials research datasets.
Have a great read. ☕
📰 Top picks of the week
GitHub Launches Free Copilot for 150M Developers in VS Code
GitHub celebrates a milestone of 150 million developers by introducing GitHub Copilot Free, now available in VS Code. Users can access up to 2,000 code completions and 50 chat messages monthly by signing in. Copilot Free supports both Anthropic's Claude 3.5 and OpenAI�s GPT-4o models for coding assistance. Students and educators continue to enjoy unlimited access to Copilot Pro accounts.
AWS Launches High Memory EC2 U7inh Instance with 32 TB Memory and 1920 vCPUs
AWS has launched the new Amazon EC2 U7inh instance designed for large in-memory databases, built with HPE. This instance supports 32 TB of memory and 1920 vCPUs, providing the highest compute performance in the AWS cloud. It is certified for SAP workloads, enabling seamless migration from HPE hardware to AWS. The U7inh instance is currently available in the US East (N. Virginia) region.
AWS Launches Stable Diffusion 3.5 Large for Advanced Image Generation
AWS now offers Stable Diffusion 3.5 Large in Amazon Bedrock, enabling high-quality image generation from text prompts. This model, with 8.1 billion parameters, excels in diverse styles and efficient prompt adherence. It supports various creative applications such as concept art, visual effects, and advertising. Users can access and test this model via the Amazon Bedrock console for enhanced image generation workflows.
Microsoft Moves to Delete Passwords for 1 Billion Users Amid 200% Rise in Attacks
Microsoft plans to delete passwords for one billion users as password-related attacks surge by nearly 200%. The company reports blocking 7,000 password attacks per second and emphasizes the advantages of adopting passkeys. Passkeys improve user experience, enhance security, and reduce the risk of forgotten passwords. The goal is to encourage users to embrace this passwordless future, which is significantly faster and more successful than traditional password methods.
Phi-4: Microsoft Newest Small Language Model Specializing in Complex Reasoning
Phi-4 is a 14B-parameter small language model that excels at complex math reasoning and general language tasks. It outperforms larger models on math benchmarks and is available on Azure AI Foundry now, with a Hugging Face release coming soon. Built with high-quality data and responsible AI tools, Phi-4 enables safe, reliable AI solutions for developers.
Google Introduces Veo 2: High-Quality AI Video Generation Up to 4K
Veo 2 is an advanced video generation model by Google that creates high-quality videos up to 4K resolution. It excels in realism and motion accuracy, simulating real-world physics and various visual styles. Users can achieve greater control over camera angles and movements with precise interpretations of instructions. The model offers enhanced detail and reduced artifacts compared to other AI video models.
Google Unveils Agentspace: Empowering Enterprises with AI for Enhanced Productivity
Google has launched Google Agentspace, designed to enhance enterprise productivity using AI. It combines Gemini's advanced reasoning, Google-quality search, and enterprise data for efficient task completion. Key features include NotebookLM for interacting with data, a multimodal search agent for information discovery, and automation through expert agents. Employees can now access complex insights and make informed decisions more easily.
Hugging Face Launches ModernBERT: Enhanced Encoder Models with 8192 Seq Length and 139M Params
The article introduces ModernBERT, an advanced set of encoder-only models that outperform older versions like BERT. It features an impressive sequence length of 8192, superior processing speed, and better performance in various tasks. ModernBERT offers two model sizes: base (139M params) and large (395M params), and is available in the latest version of the Hugging Face transformers library. Users can efficiently fine-tune ModernBERT for tasks such as classification and question answering.
Hugging Face Unveils Bamba-9B Model with 2.5x Throughput Improvement
The Bamba-9B model, a hybrid Mamba2 model developed by IBM and several universities, improves inference efficiency by offering 2.5x throughput and 2x latency speedup over standard transformers. It is designed to tackle KV-cache memory bottlenecks and is fully open for community use in various frameworks. The model includes tuning and training scripts, enabling further experimentation and improvement. Future work aims to enhance its capabilities through collaborative efforts.
Hugging Face Launches LeMaterial: A Unified Dataset of 6.7 Million Entries for Materials Research
LeMaterial is an open-source project by Entalpic and Hugging Face aimed at advancing materials discovery through a unified dataset. The release of LeMat-Bulk combines 6.7 million entries from major materials databases like Materials Project and OQMD. This initiative addresses challenges in dataset integration and biases, enabling faster research and development of new materials. It leverages machine learning to enhance the discovery of novel compounds and improve efficiency in materials science.
❤️ You might also like
🏁 Enjoy this newsletter?
Forward it to a friend, and let them know they can subscribe here.