featherless-ai/RWKV-small-classifier

Hugging Face
TEXT GENERATIONPricing:Input $0.03 / Output $0Concurrent Unit Cost:1Model Size:2.9BQuant:BF16Context Size:32k Classification model Architecture:Transformer Warm

The featherless-ai/RWKV-small-classifier is a 2.9 billion parameter RWKV-based language model designed for structured classification tasks. It leverages the Simple Jev classifier API to convert text, conversations, or structured data into structured decisions like choice selection, rubric scoring, or proposition judgment. This model is optimized for applications requiring intent detection, content moderation, or routing, providing confidence scores and probabilities for its classifications.

Loading preview...

Overview

The featherless-ai/RWKV-small-classifier is a 2.9 billion parameter model built on the RWKV architecture, specifically designed for structured classification tasks. It operates via the Simple Jev classifier API on Featherless, allowing developers to transform unstructured context (text, chat history, or structured data) into actionable, structured decisions. This model is currently in beta on Featherless developer plans.

Key Capabilities

  • Structured Decision Making: Converts input into JSON outputs for direct application use.
  • Multiple Question Types: Supports choice (selecting from named candidates), score (rating against an ordered rubric), and noul (yes/no proposition judgment).
  • Confidence and Probabilities: Provides confidence scores and probabilities for each classification, aiding in result interpretation.
  • Text and Chat History Context: Processes string, JSON object, array, or chat message history as input.
  • Independent Question Evaluation: Allows multiple questions to be evaluated against shared context, with each question assessed independently.

Good For

  • Support Routing: Automatically direct customer inquiries to the correct department.
  • Intent Detection: Identify user intent in conversations or text inputs.
  • Content Moderation: Classify content for appropriateness or policy adherence.
  • Relevance Scoring: Rate the relevance of items based on specific criteria.
  • Agent Action Selection: Determine an AI agent's next action based on current context.

Unlike some other classifier models, this RWKV variant is text-only, meaning it does not support image or vision inputs.