Overview & result
love jev, but upset it - isn't open source? - it lack vision capability? We fixed all of that, introducing SimpleJev.ai A fully open source library which takes any HF model and Jev-ify it, with an API endpoint Now on github, and live in production at @FeatherlessAI
How Jev fits in the loop
- You send shared context and a set of questions.
- The server applies the model's chat template; questions about the same context share most of the prompt.
- It reads the next-token logits from the prefill step to score the predefined answer labels, with no decode loop.
- The server builds the JSON response from those scores; the model never generates JSON.
How to reproduce
- Inspect the original showcase and source material at https://github.com/featherless-ai/simple-jev
- Verify the bounded prompt and input candidate schema configured for Jev
- Benchmark decision latency and classification accuracy against baseline models
Why this build matters
It gets structured classification and scoring from compatible open Hugging Face models without training a separate classifier head.
Limitations
- The current implementation runs locally with Hugging Face Transformers and PyTorch.
- The public demo API has a 2k-token context limit and a 2 requests per second rate limit.
- Questions per request default to 100, with a schema maximum of 256.