Meta’s Llama model had been initially released for research purposes, but its weights had ultimately been leaked, enabling it to become the basis of dozens of open-source models. But Meta has now released a new version of Llama that’s explicitly open-source and open for commercial use.
Meta has released Llama v-2, which is open source and available for commercial use. “Llama-v2 is open source, with a license that authorizes commercial use!” announced Meta’s Chief AI Scientist Yann LeCun. “This is going to change the landscape of the LLM market. Llama-v2 is available on Microsoft Azure and will be available on AWS, Hugging Face and other providers Pretrained and fine-tuned models are available with 7B, 13B and 70B parameters,” he added.
“We develop and release Llama 2, a collection of pretrained and fine-tuned large language models (LLMs) ranging in scale from 7 billion to 70 billion parameters,” says Llama 2’s paper. “Our fine-tuned LLMs, called Llama 2-Chat, are optimized for dialogue use cases. Our models outperform open-source chat models on most benchmarks we tested, and based on our human evaluations for helpfulness and safety, may be a suitable substitute for closedsource models. We provide a detailed description of our approach to fine-tuning and safety improvements of Llama 2-Chat in order to enable the community to build on our work and contribute to the responsible development of LLMs,” it adds.
Llama 2 has 7B, 13B and 70B parameters. It’s been trained on 40% more data than Llama 1. It had 2 trillion pretraining tokens, and has a context length of 4096 tokens.
Llama’s release could end up upturning the entire LLM space. OpenAI has resolutely kept its models closed-source, and Google and Anthropic AI’s models are closed-source too, but Llama-2 is supposedly comparable to GPT-3.5, and free and open to commercial use to boot. A free and open-source model that’s extremely capable will help researchers tinker with it and improve it in all kinds of ways, and it’s possible that these fine-tuned models could surpass GPT-4 in capability. It’s still early days — the coming weeks will allow researchers to figure out the capabilities of the model — but Meta’s launch of Llama-2 has thrown the LLM field wide open.