Llama 3.1 405B Instruct
ConditionalMeta · Llama · Released 2024-07
405BText
Meta's first frontier-scale open-weight model, competitive with closed frontier models at release but out of reach for anyone without a GPU cluster.
Strengths
- +Closed-frontier-competitive reasoning and knowledge
- +Distillation source for Llama 3.1/3.2 smaller models
- +128K context window
Limitations
- -Requires multi-node, multi-GPU serving for full-precision inference
- -Very expensive to fine-tune or self-host at scale
License
Llama 3.1 Community License
free with acceptable-use or scale conditions - read it
Hardware
needs multi-GPU or datacenter hardware to run well
Links
Stats
meta-llama/Llama-3.1-405B-Instruct— downloads·— likes
via Hugging Face