nobodywho
nobodywho is an open-source project focused on localizing the decision step of LLMs. Its decision model scores candidate answers directly and returns a probability distribution, replacing the conventional generative JSON output and parsing pipeline. It is suited to scenarios such as routing selection and safety filtering.
Timeline (last 90 days)
The nobodywho project has introduced Decision Models, which do not generate tokens but instead score candidate answers directly, returning the probability and confidence for each option.
Decision Models are invoked through Python's Decision or DecisionAsync interface, and the model runs locally without network requests.
Decision Models use the laya model under the hood, which is 449 MB in size and licensed under Apache-2.0.
Currently, Decision Models only support local operation, and the 449 MB model size may still be a burden for resource-constrained environments.
The nobodywho project has been open-sourced on GitHub, the documentation provides Python interface examples, and the repository address is https://github.com/nobodywho-ooo/nobodywho.
All Facts (5)
The nobodywho project has introduced Decision Models, which do not generate tokens but instead score candidate answers directly, returning the probability and confidence for each option.
50%UnverifiedDecision Models are invoked through Python's Decision or DecisionAsync interface, and the model runs locally without network requests.
50%UnverifiedDecision Models use the laya model under the hood, which is 449 MB in size and licensed under Apache-2.0.
50%UnverifiedCurrently, Decision Models only support local operation, and the 449 MB model size may still be a burden for resource-constrained environments.
50%UnverifiedThe nobodywho project has been open-sourced on GitHub, the documentation provides Python interface examples, and the repository address is https://github.com/nobodywho-ooo/nobodywho.
50%