Llama 2

Llama 2 is a collection of pre-trained and fine-tuned generative text models ranging in scale from 7 billion to 70 billion parameters published in July 2023 by Meta. It was released under a special Llama 2 license.

Llama 2 is an auto-regressive language model that uses an optimized transformer architecture. The tuned versions use supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to align to human preferences for helpfulness and safety.

Llama 2 was pre-trained on 2 trillion tokens of data from publicly available sources. The fine-tuning data includes publicly available instruction datasets, as well as over one million new human-annotated examples.

The pre-training data has a cutoff of September 2022, but some tuning data is more recent, up to July 2023.

Supported variants

Airtrain supports three variants of the Llama 2 model.

llama-2-7b-chat

Llama-2-7b-chat is a 7B parameter model optimized for dialogue use cases. It outperforms many similarly sized open-source chat models on most benchmarks. In human evaluations for helpfulness and safety, it's on par with some popular closed-source models like ChatGPT and PaLM.

See the full model card on HuggingFace.

llama-2-13b-chat

Llama-2-13b-chat is a 13B parameter model optimized for dialogue use cases. It outperforms many similarly sized open-source chat models on most benchmarks. In human evaluations for helpfulness and safety, it's on par with some popular closed-source models like ChatGPT and PaLM.

See the full model card on HuggingFace.

llama-2-70b-chat

Llama-2-70b-chat is a 70B parameter model optimized for dialogue use cases. It outperforms many open-source chat models on most benchmarks. In human evaluations for helpfulness and safety, it's on par with some popular closed-source models like ChatGPT and PaLM.

See the full model card on HuggingFace.