FireLLaVA
Fireworks AI
Llama 2 Community
About
FireLLaVA is a family of vision-language models (VLMs) innovatively developed by Fireworks AI. Built on the LLaVA architecture, these models are celebrated for their open-source nature, released under the Llama 2 Community License, making them the first commercially permissive LLaVA models available 58. Unlike previous VLMs such as the original LLaVA, which faced commercial restrictions due to proprietary training data, FireLLaVA models utilize open-source instruction-following data, achieving performance on par with or surpassing prior benchmarks 58. Notably, the FireLLaVA-13b model, trained in December 2023 and available on Hugging Face, is tailored for single-image inputs while also supporting multi-image and multi-prompt generation 8.