Unpack the Latest Llama 4 Herd
summary
Claims concrete performance and deployment details: two new models with 17 billion active parameters each—Scout is distilled to run on a single NVIDIA H100 and offers an industry-leading 10 million token context window, Maverick uses 128 experts and supports a 1 million token window; both ingest up to five images for multi-image understanding and 12 languages, are distilled from a 288B Behemoth still in training, power demos that build a gallery site in under five minutes, and are available via download, Hugging Face, or the Llama API.
AI-written summary · about
chapters (9)
- 0:00Llama 4 introduction and evolution
- 1:57Llama models real-world use cases
- 3:25Developer feedback on Llama 4
- 4:31Llama 4 Scout and Maverick features
- 6:21Llama 4 context windows and multilingual support
- 8:12Llama 4 multi-image understanding
- 9:20Llama 4 demo
- 9:41Llama 4 Cookbook and setup
- 10:45Accessing and using Llama 4
related talks
16mLlama 4
48mBringing Llama 3 to Life
17mThe State of the Llama Ecosystem
31mBuilding LLaMA: Meta's Director of GenAI
38mMeta's Roadmap for Full Stack AI
12mKeynote: The Llama Ecosystem: Past, Present and Future
44mPimp My Llama - How to Customize LLMs for your Business Case
52mHW, SW, Performance and Costs for Llama-2 70b and Mixtral 8x7b LLM Inference with Low...