Meta Open-Sources Llama 3.2, Adding a 70-B Model and a $150M Fund

Translated for your language. Read the original.

AI-assisted, human-reviewed.

GyaanSetu Editorialjana1min read

Meta Open-Sources Llama 3.2, Adding a 70-B Model and a $150M Fund

Meta Releases Llama 3.2 Open Source

Meta opened Llama 3.2 to the public, adding a 70 billion-parameter model. You can pull the code from GitHub or Hugging Face.

The model was trained on a 2.5 trillion-token dataset. Meta ships three sizes to suit different workloads:

8 B
30 B
70 B

All three run on modern GPUs and LPU chips. A new architecture cuts compute demand by 22 %, which translates into lower costs and less energy use.

Why this matters for you

Lower costs – Startups and developers can now run large models without massive budgets.
Better performance – Benchmarks show Llama 3.2 outpaces several open-source rivals on reasoning and multilingual tasks.
Commercial freedom – The Apache 2.0 license lets you deploy the model in a business without paying royalties.
Hardware speed – The model runs 1.8 × faster on chips like NVIDIA H100 and Groq LPU.

Meta is pouring $150 million into a fund that backs projects built on Llama 3.2, pushing AI development out of closed labs and into the community.

Source: https://dev.to/techpulse01239/meta-ai-unveils-llama-32-open-source-release-70b-model-3bhb

Optional learning community: https://t.me/GyaanSetuAi