November 29 The Perplexity team presented PPLX Online LLMS project-two new PPLX-7B- models Online and PPLX-70B-Online, which differ from traditional LLM models (Large Language Models), the ability to use information from the Internet, which allows you to provide relevant, useful answers based on real facts. Models are available through the API PPLX and can be used through the Perplexity Labs site.
an example of the response of the model PPLX-70B-online
PPLX online llm features include:
- relevance of information: models are able to provide the most relevant information, which is an advantage over traditional LLM models;
- The use of open models: PPLX models are built on the basis of the basic models of Mistral-7B and Llama2-70b.
- Advanced search technology: Own search, indexing and scanning sites allows models to use the most relevant and relevant information;
- Subtle setting (Fine-Tuning): Models are adjusted for the efficient use of excerpts from sites (snippets) for the formation of answers.
The effectiveness of PPLX models was evaluated through a series of tests, including comparisons with other popular models such as the GPT-3.5-Turbo-1106 from Openai and Llama2-70b-Cheat from META* AI. The results showed that the PPLX-7B-ONLINE and PPLX-70B-ONLINE models are preferable by the criteria of freshness, factual and integrity of answers.
Comparison of PPLX-7B-ONLINE and PPLX-70B-ONLINE models with LLAMA22 70B and GPT-3.5
The public version of the API PPLX was also announced, along with the new pricing system based on use. Pro users will receive a monthly loan of $ 5 for use in the API.
*META and its products are recognized as extremist, their activities are prohibited in the territory of the Russian Federation.