For Featherless, we initially attempted to use it, but the free plan did not contain any frontier models (models were capped to 15 billion parameters), and also featherless used a subscription-based credit system rather than a per-token based system, which would prevent this project from being scalable. Due to these problems with featherless, we pivoted to NanoGPT (https://nano-gpt.com/), which was a model router, with pricing per token (which would be cheaper in the long run). It's SDK was the same as the standarized OpenAI SDK, thus making the pivot easy.
Log in or sign up for Devpost to join the conversation.