The platform is able to run many models due to its custom inference stack, which can dynamically swap out models in less than a second. They use FP8 quantization, which maintains output quality while improving inference speeds. Featherless aims to make Open Source AI accessible to everyone and the revenue from the platform supports their work in training and scaling open source AI models. They encourage users to reach out to their favorite model creators to support them directly.
Key takeaways:
- Featherless is an AI model provider that offers subscribers access to a growing library of Hugging Face models, with new models added weekly.
- They offer two pricing plans, Feather Pro at $10 a month and Feather Premium at $25 a month, with different benefits and concurrency limits.
- Featherless prioritizes privacy, not storing any chat, prompts or completions logs from users.
- They support LLaMA-3-based models, including LLaMA-3 and QWEN-2, and plan to add more architectures to their supported list soon.