|
gptkbp:instanceOf
|
large language model
|
|
gptkbp:architecture
|
gptkb:transformation
|
|
gptkbp:author
|
gptkb:Meta_AI_Research
|
|
gptkbp:availableOn
|
gptkb:Hugging_Face
gptkb:Meta_AI_website
|
|
gptkbp:benchmarkPerformance
|
approaches GPT-4 (in some tasks)
competitive with GPT-3.5
outperforms Llama 2
|
|
gptkbp:context
|
128K tokens (for some variants)
8K tokens
|
|
gptkbp:developedBy
|
gptkb:Meta_AI
|
|
gptkbp:fineTuning
|
supported
|
|
https://www.w3.org/2000/01/rdf-schema#label
|
Llama 3
|
|
gptkbp:license
|
custom commercial license
|
|
gptkbp:mobileAppAvailable
|
gptkb:Amazon_SageMaker
gptkb:Google_Cloud
gptkb:Microsoft_Azure
gptkb:Meta_API
|
|
gptkbp:modelCardURL
|
https://llama.meta.com/llama3
|
|
gptkbp:notableFeature
|
multilingual support
better instruction following
commercial use allowed
improved reasoning
improved safety alignment
open weights
supports RAG
supports code generation
supports function calling
supports long context windows
|
|
gptkbp:notablePublication
|
gptkb:Llama_3:_Open_Foundation_and_Instruction_Models
https://llama.meta.com/llama3
|
|
gptkbp:notTrainedOn
|
Meta user data
|
|
gptkbp:openSource
|
yes
|
|
gptkbp:parameter
|
400B (rumored unreleased)
70B
8B
|
|
gptkbp:predecessor
|
gptkb:Llama_2
|
|
gptkbp:quantizedVersions
|
available
|
|
gptkbp:releaseYear
|
2024
|
|
gptkbp:supports
|
gptkb:RAG_(retrieval-augmented_generation)
English
multilingual
tool use
function calling
|
|
gptkbp:trainer
|
publicly available data
Meta-licensed data
data from partners
|
|
gptkbp:uses
|
research
code generation
text generation
chatbots
|
|
gptkbp:bfsParent
|
gptkb:transformation
gptkb:GPT-3
gptkb:Llama_model
|
|
gptkbp:bfsLayer
|
5
|