Facts about Llama 2

1. 07

   Llama 2-Chat, the conversational variant of Llama 2, underwent supervised fine-tuning with feedback from human annotators who ranked model responses to improve dialogue quality and safety.

   Llama 2May 14sharetrainingaialignment
2. 06

   In competitive evaluations, Llama 2 matched or exceeded GPT-3.5 performance on 24 out of 30 standard reasoning and coding benchmarks despite having significantly fewer parameters.

   Llama 2May 14shareperformanceaicomparison
3. 05

   Llama 2's context window of 4,096 tokens allows it to process roughly 3,000 words in a single conversation before requiring new context.

   Llama 2May 14sharetechnicalmeasurementlanguage
4. 04

   Open-source developers can commercially use Llama 2 without requesting permission from Meta, distinguishing it from many competing large language models with restrictive licenses.

   Llama 2May 14sharelicensingcommercialopen-source
5. 03

   Llama 2's safety training incorporated constitutional AI principles, using a set of rules to guide the model away from generating harmful content.

   Llama 2May 14sharesafetyaitraining
6. 02

   Through instruction tuning on 27,540 annotated examples, Llama 2 achieved competitive performance with closed-source models like GPT-3.5 across multiple benchmarks.

   Llama 2May 14shareartificialintelligencetrainingperformance
7. 01

   Meta released Llama 2 in July 2023 with versions ranging from 7 billion to 70 billion parameters for commercial use.

   Llama 2May 14sharetechnologymeasurementrelease