Typed decisions, one request, up to 32K tokens of context
Send a state (a document, an email, a JSON object) and any number of Choice, Noul and Score
questions. Bobcat returns a probability for every option you offered and nothing else: no generated text, no
options it made up. The state is read once for all the questions in a request. Here Bobcat or Bobcat Flash
runs inside this Space on a shared GPU in FP8: no key and no sign-up.
Model
Shared GPUConnecting to this Space...
State
Questions
Request body
The same System One request a Bobcat server takes; this page hands it to the Space's decide endpoint through the Gradio client.
Response (the closed JSON, as received)
No response yet.
Serve the same models on your own GPU from their Hugging Face repositories
(the quickstart of the model cards), then send this request with the official typesafe-sdk.
A --local server accepts any key: export BOBCAT_API_KEY=local.