shangyangwu2002/SCAD

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Oct 2, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

SCAD-4B is a 4 billion parameter text-only causal language model developed by shangyangwu2002, fine-tuned from Qwen3-4B with a 32768 token context length. It is specifically designed for structured credit assignment and distillation in long-horizon agents. This model is optimized for agentic tasks, leveraging a dedicated retrieval corpus and associated code for agent prompts, search tools, and evaluation.

Loading preview...

SCAD-4B Overview

SCAD-4B is a 4 billion parameter text-only causal language model, fine-tuned by shangyangwu2002 from the Qwen3-4B architecture. It is specifically developed for applications involving Structured Credit Assignment and Distillation (SCAD) in long-horizon agents, offering a 32768 token context window. This model is designed to work in conjunction with a dedicated retrieval corpus and a specialized code repository that provides agent prompts, search tools, and evaluation mechanisms.

Key Capabilities

  • Agentic Task Optimization: Fine-tuned for complex, long-horizon agent operations.
  • Structured Credit Assignment: Facilitates advanced credit assignment within agent frameworks.
  • Distillation for Agents: Supports distillation processes relevant to agent learning.
  • Text-Only Processing: Focuses on natural language understanding and generation for agent control.
  • Extended Context Window: Utilizes a 32768 token context length for processing longer interactions and information.

Good For

  • Developing Long-Horizon AI Agents: Ideal for researchers and developers building agents that require sustained reasoning and planning over extended periods.
  • Research in Agentic AI: Suitable for exploring structured credit assignment, distillation techniques, and advanced agent architectures.
  • Applications Requiring External Knowledge Integration: Designed to integrate with a separately hosted retrieval corpus for enhanced information access.
  • Custom Agent Prompting and Tool Use: Leverages an associated code repository for implementing specific agent prompts and search tools.

Finetunes
1 models