featherless-ai/RWKV-std-classifier

Hugging Face
TEXT GENERATIONPricing:Input $0.2 / Output $0Concurrent Unit Cost:1Model Size:13.3BQuant:FP8Context Size:32k Classification model Architecture:Transformer Warm

The featherless-ai/RWKV-std-classifier is a 13.3 billion parameter RWKV-based model, developed by Featherless, specifically designed for structured classification tasks. It leverages the Simple Jev classifier API to convert text context into structured decisions, supporting choice, scoring, and yes/no propositions. This text-only model excels at applications like intent detection, content moderation, and routing, providing JSON outputs for direct application use.

Loading preview...

RWKV-std-classifier: Structured Decision-Making with Simple Jev

The featherless-ai/RWKV-std-classifier is a 13.3 billion parameter model, part of the RWKV family, developed by Featherless. It is specifically engineered to transform textual context into structured, actionable decisions using the Simple Jev classifier API. This model is optimized for classification tasks, providing JSON outputs that can be directly integrated into applications.

Key Capabilities

  • Structured Classification: Converts text, conversations, or structured data into defined answers, scores, or propositions.
  • Multiple Question Types: Supports three primary classification types:
    • choice: Selects a candidate from a predefined list (e.g., routing support tickets).
    • score: Rates context against an ordered rubric (e.g., assessing urgency or relevance).
    • noul: Determines support for a yes/no proposition (e.g., detecting refund requests).
  • Text-Only Input: Processes text and chat history; it does not support image or vision inputs.
  • JSON Output: Provides structured JSON responses including the chosen answer, confidence levels, and probabilities for each candidate.

Ideal Use Cases

This model is particularly well-suited for scenarios requiring automated decision-making based on textual input, such as:

  • Support Routing: Automatically directing customer inquiries to the correct department.
  • Intent Detection: Identifying user intentions in conversations or queries.
  • Content Moderation: Classifying content for compliance or appropriateness.
  • Relevance Scoring: Rating the relevance of information to a specific query.
  • Agent Action Selection: Guiding AI agents to choose the next best action based on context.

For production use, the model integrates with the Featherless API, offering higher limits and dedicated support. Further technical details on the underlying prompt format and API can be found in the Simple Jev GitHub repository.