Amy Angelini

*pinned bio: 9 times out of 10 I'm tweeting about my grandma. Or my cat. * ^in memory | thoughts are my own, but I think that goes without saying

[VentureBeat @VentureBeat]

IBM debuts open source Granite 3.0 LLMs for enterprise AI https://t.co/TjwQ4F8uqN

[elvis @omarsar0]

IBM presents Granite 3.0, lightweight foundation models ranging from 400 million to 8B parameters.

Supports coding, RAG, reasoning, and function calling, focusing on enterprise use cases, including on-premise and on-device settings.

The technical report contains detailed discussions on how they collect synthetic datasets for code, reasoning, RAG, tool use, and more.

"Granite 3.0 language models demonstrate strong performance across a battery of academic benchmarks for language understanding, reasoning, coding, function calling, and safety"

The release includes pre-trained and post-trained versions of all Granite 3.0 models under a permissive Apache 2.0 license.

Very strong release by the Granite Team at IBM.

[Rohan Paul @rohanpaul_ai]

IBM just dropped 8B and 2B param models that outperformed similarly sized Llama-3.1 8B in Hugging Face's OpenLLM Leaderboard (v2)

- Third generation of Granite LLM series
- 8B and 2B parameter dense decoder-only models
- Trained on 12+ trillion tokens across 12 languages and 116 programming languages
- Matches/exceeds performance of similar open models
- Apache 2.0 license with disclosed training data and methods
- Available on IBM watsonx platform and through partners
- Trained using 100% renewable energy

🏆 Performance Benchmarks

- Rivals Meta and Mistral AI models on academic benchmarks
- Leads on RAGBench for enterprise tasks
- Excels on cybersecurity benchmarks
- Outperforms peers on function calling benchmarks
- Industry-leading robustness on adversarial prompt benchmark AttaQ

⚡ Speed and Efficiency Innovations

The Granite 3.0 release also includes additional inference-efficient offerings: mixture of experts (MoE) models and a speculative decoder for accelerated inference.

- Mixture of Experts (MoE) models: 3B-A800M and 1B-A400M
- High efficiency for on-device, CPU servers, low-latency use cases
- Speculative decoding model accelerates inference by 220%
- Innovations in token conditioning and two-phase training

🛠️ Getting Started

- Guides in Granite Snack Cookbook on GitHub
- Granite model playground for experimentation
- Demos and tutorials on IBM. com
- Access through IBM watsonx. ai and partners

@IBM

[ollama @ollama]

Today, IBM and Ollama announces partnership to bring Granite 3.0 models! 😍

2 architectures:

Dense:

2B:
ollama run granite3-dense

8B:
ollama run granite3-dense:8b

Mixture of Experts:

1B:
ollama run granite3-moe

3B:
ollama run granite3-moe:3b

Blog post:
https://t.co/xLqCexAaHr

[Amy Angelini @AimAng]

we're out here!

[Steve Lohr @SteveLohr]

Trying to make generative AI safe for corporate America. IBM's approach -- assume customers' legal risk and publish the data sets.
https://t.co/AUAlewHM0Z

[Amy Angelini @AimAng]

@GoodsCapital @Even_Moderation Wow is @Even_Moderation about to launch his 2024 presidential bid?

[Amy Angelini @AimAng]

@Cdubsz This has been severely overlooked

[Amy Angelini @AimAng]

@eriangelini @CruzKayne LOL 20 percent tip? 30? here’s my credit card … keep it!

[IBM News @IBMNews]

Breaking: IBM acquires leading data & analytics software provider for environmental performance management @Envizi. IBM's AI portfolio now gives clients a single source of truth for analyzing hundreds of factors related to environmental impact. Read more: https://t.co/npRZlw4Lji

[Amy Angelini @AimAng]

I just feel like “thank you” is one phrase we don’t need to abbreviate

[The New Yorker @NewYorker]

Sudowrite is an application that harnesses the artificial-intelligence program GPT-3 to generate text and even mimic the literary style of writers. “Whatever field you are in, if it uses language, it is about to be transformed.” https://t.co/xhM1nanP6H

[Good Morning America @GMA]

In the face of adversity, problem solvers around the world rise up to answer the call.

Join the #CallforCode Global Challenge with @IBMDeveloper today to build social impact solutions to fight climate change. https://t.co/n72Br2XKMi

[Erin Angelini @eriangelini]

we've all been there

[Amy Angelini @AimAng]

@UberFacts @CMaster_HYHY you’re special

[Erin Angelini @eriangelini]

"...Alzheimer’s is just the beginning. People with a wide variety of neurological illnesses have distinctive language patterns that, investigators suspect, may serve as early warning signs of their diseases." @ginakolata @nytimes @IBMResearch https://t.co/iMaDZtbRIa

[Amy Angelini @AimAng]

big props to the avocado for staying *relevant*

[UberFacts @UberFacts]

Avocados were eaten by giant ground sloths 35 million years ago

[Amy Angelini @AimAng]

I need a talking point on why my eye cream needs to be different than my rest of face lotion

[Amy Angelini @AimAng]

@Cdubsz I have some logistical questions

[Amy Angelini @AimAng]

@Cdubsz she doesn’t want our kind

[Amy Angelini @AimAng]

can we really blame the moon?

[UberFacts @UberFacts]

The moon is slowly drifting away from earth at a rate of about 1.5 inches per year