• AIcodeX
  • Posts
  • Apple going open source with Tiny models

Apple going open source with Tiny models

Author

AI codeX
April 25, 2024

Join Now  | Advertise

WELCOME, AI LOVERS 

In today’s AI Decode →

  • Apple tiny open source models

  • Snowflake shoots for enterprise

  • GitHub Copilot Competitor with $252 M Funding

  • 6 research and 5 AI tools

  • More AI & tech Stories

Apple AI
  Apple going open source with Tiny models

Apple has quietly released OpenELM, a new family of open-source large language models designed to run efficiently on a single device without requiring cloud connectivity. Just as Google, Samsung and Microsoft push their efforts with genAl on PCs and mobile devices

Decoding the Details :

  • OpenELM consists of 8 models (4 pre-trained, 4 instruction-tuned) ranging from 270 million to 3 billion parameters for on-device text generation tasks.

  • The models were pre-trained on 1.8 trillion tokens from public datasets like Reddit, Wikipedia, and arXiv using Apple's new CoreNet library. •

  • Performance is respectable but not bleeding-edge, with the 450 million instruct variant showing promising results across benchmarks. •

  • While smaller than cutting-edge models, OpenELM aims to facilitate open research in efficient on-device generative Al capabilities.

snowflake AI
❄️Snowflake shoots for enterprise with Arctic AI model

Snowflake unveils Arctic, an open and efficient LLM optimized for complex enterprise workloads, aiming to democratize data and enhance AI capabilities for businesses.

 Decoding the Details :

  • Arctic uses a Dense Mixture of Experts architecture to deliver targeted AI performance for SQL and code generation, and instruction following, with minimal compute usage.

  • The model competes well with industry benchmarks, offering nearly equivalent performance to leading models from Databricks, Meta, and Mistral.

  • Snowflake emphasizes cost-effectiveness, with Arctic requiring significantly less compute power for training compared to other models.

  • Arctic’s weights and code are available under the Apache 2.0 license on platforms like Hugging Face, with additional resources provided to aid in model training and development.

Augment
🦄Augment Emerges as GitHub Copilot Competitor with $252 Million in Funding

AI-Powered Coding Platform Augment Emerges from Stealth with $252 Million in Funding. Augment, founded by ex-Microsoft developer Igor Ostrovsky, aims to revolutionize the AI-driven coding

Decoding the Details :

  • Augment secured funding from notable investors, including Eric Schmidt and top VCs, at a near-unicorn valuation

  • The platform utilizes fine-tuned, industry-leading open AI models to enhance software development processes

  • Augment plans to offer standard SaaS subscriptions and will reveal pricing and other details closer to its GA release

  • The startup faces competition from tech giants and numerous other AI coding assistant startups’s in an increasingly crowded market

 🧰 AI Tools

  • Anon - The toolkit for building AI integrations for sites without APIs.

  • Greetio - Send eCards with personalized AI-assisted poems

  • Looka AI: AI-powered platform to design a logo

  • Papermark AI: Your open-source AI document assistant. chat with documents

  • HomeStage: Upload a picture and instantly add virtual furniture to any room

🧬New Research
arXiv is a free online library where researchers share pre-publication papers.

  🌐More Stories

Nvidia CEO Jensen Huang personally delivers first DGX H200 to OpenAI

Nvidia acquires AI workload management startup Run:ai for $700M, sources say

Nooks raises $22M to empower sales reps with its AI-powered call platform

UK probes Amazon and Microsoft over AI partnerships with Mistral, Anthropic and Inflection

Biden signs TikTok ‘ban’ bill into law, starting the clock for ByteDance to divest it

Thanks for reading.

Catch you next time